Quadruped Robots: Bridging Mechanical Design, Control, and Applications

Abstract

Quadruped robots have emerged as a prominent field of research due to their exceptional mobility and adaptability in complex terrains. This paper presents an overview of quadruped robots, encompassing their design principles, control mechanisms, perception systems, and applications across various industries. We review the historical evolution and technological milestones that have shaped quadruped robotics. To understand their impact on performance and functionality, key aspects of mechanical design are analyzed, including leg configurations, actuation systems, and material selection. Control strategies for locomotion, balance, and navigation are all examined, highlighting the integration of artificial intelligence and machine learning to enhance adaptability and autonomy. This review also explores perception and sensing technologies that enable environmental interaction and decision-making capabilities. Furthermore, we systematically examine the diverse applications of quadruped robots in sectors including the military, search and rescue, industrial inspection, agriculture, and entertainment. Finally, we address challenges and limitations, including technical hurdles, ethical considerations, and regulatory issues, and propose future research directions to advance the field. By structuring this review as a systematic study, we ensure clarity and a comprehensive understanding of the domain, making it a valuable resource for researchers and engineers in quadruped robotics.

1. Introduction

With the rapid advancement of technology, robotics has emerged as one of the most influential fields in the 21st century [1]. In particular, quadruped robots have gained significant attention due to their exceptional mobility and adaptability in complex terrains. Inspired by the locomotion of four-legged animals, these robots integrate mechanical design [2,3], control theory [4], computer vision [5,6,7], and artificial intelligence [8,9,10] to perform tasks in environments inaccessible or hazardous to humans.

Quadruped robots [11] hold immense potential across various applications, including military reconnaissance [12,13], disaster response [14,15,16,17,18], industrial inspection, agricultural operations, and service industries. For example, they [14] can navigate through earthquake rubble to search for survivors or inspect equipment in intricate industrial settings. These scenarios demand high levels of stability, flexibility, and autonomy from robots.

Despite considerable progress, research and deployment involving quadruped robots still faces numerous challenges. Complex mechanical structures, efficient motion control algorithms, reliable perception systems, and energy management are all critical issues that must be addressed. Recent advancements in sensor technologies, materials science, and artificial intelligence have significantly enhanced the performance and capabilities of quadruped robots, opening up new avenues for research and application.

Several existing surveys [11,19,20,21,22,23,24,25,26,27,28,29,30,31] have explored different aspects of quadruped robots (see

Table 1. Summary of quadrupedal robotics reviews (sorted by year).

Ref.	Year	Summary
[19]	2020	Comprehensively reviews high-dynamic motion capabilities and structural and control technologies in legged robots for versatile tasks.
[20]	2021	Explores quadrupedal robots’ historical evolution and modern capabilities, including mobility planning and environmental sensing.
[21]	2021	Discusses bio-inspired designs, sensors, and algorithms in robotics, emphasizing challenges and innovations in adaptive and efficient behaviors.
[22]	2021	Examines driving modes of quadrupedal robots, comparing hydraulic and motor-driven systems for task-specific efficiency.
[23]	2022	Analyzes quadrupedal robot design, control, and operational advancements, including task adaptability and multi-functional mechanisms.
[24]	2022	Reviews localization, mapping, and trajectory planning for quadrupedal robots in vineyards, highlighting challenges in unstructured terrains.
[25]	2022	Provides a comparative anatomical study of quadrupedal robots and animals, emphasizing structural and functional differences to inspire future robot designs.
[26]	2023	Analyzes mechanisms, energy optimization, gait planning, and intelligent sensing, emphasizing practical efficiency and future directions.
[27]	2023	Systematically reviews intelligent control methods for smooth motion in legged robots, focusing on planning, stability, and learning models.
[28]	2023	Comprehensively reviews quadrupedal mobile robots, focusing on key mechanisms, gait planning, dynamic control, and future applications.
[29]	2024	Details the design, locomotion, and potential applications of quadrupedal robots, with a focus on adaptability to rugged terrains.
[30]	2024	Focuses on integrating wearable electronics into quadrupedal robots to enhance sensing, adaptability, and operational capabilities.
[11]	2024	Highlights advancements, challenges, and future perspectives in quadrupedal robots, focusing on control paradigms, energy efficiency, and cognitive capabilities.
[31]	2024	Discusses challenges, control techniques, and emerging technologies for next-generation legged robots, focusing on robust navigation, adaptability, and energy efficiency.

). Some of these are focused on locomotion control algorithms, mainly emphasizing traditional control methods. Others discuss the prospects of quadruped robots in industrial applications, but lack detailed exposition of the mechanical design and perception systems. Recent works have concentrated on applying machine learning in quadruped robots; however, their scope is often limited to specific subfields.

Currently, there is a lack of studies that systematically integrate the design principles, control mechanisms, perception systems, and cross-industry applications of quadruped robots. This paper aims to fill the gap by providing a systematic analysis of quadruped robots in which we review their historical development, explore key technologies and innovations, compare notable case studies, discuss current challenges, and outline future research directions. This work aspires to serve as a valuable reference for researchers and engineers that can promote the further development and application of quadruped robot technology.

2. Research Methodology

This section outlines the research methodology adopted for this study, detailing the research questions, search strategy, eligibility criteria, and study selection process. The selected methodology ensures a rigorous and systematic approach to reviewing the state of quadruped robotics, focusing on mechanical design, control mechanisms, perception systems, and applications.

2.1. Research Questions

This study aims to provide a comprehensive review of quadruped robots by addressing the following research questions:

RQ1

How have quadruped robots evolved over time, and what are the major technological milestones?

RQ2

What are quadruped robots’ fundamental design principles, control mechanisms, and perception systems?

RQ3

How do quadruped robots integrate artificial intelligence and machine learning to enhance autonomy and adaptability?

RQ4

What are the key applications of quadruped robots across industries, and what challenges hinder their broader adoption?

RQ5

What are the challenges, future directions, and potential research areas for advancing quadruped robotics?

2.2. Search Strategy

To ensure a comprehensive and systematic review, an extensive search strategy was designed to gather relevant literature and data sources. This strategy involved selecting appropriate databases, defining key search terms, employing Boolean operators, tracking citations, and incorporating gray literature.

First, a selection of major academic databases was made to ensure coverage of high-quality and peer-reviewed sources. Research articles, conference papers, and review studies were collected from reputable sources such as IEEE Xplore, ScienceDirect, Springer, Web of Science, and Google Scholar. These databases provided access to a broad range of publications relevant to quadruped robotics research.

Next, a set of keywords and search terms was carefully chosen to maximize the retrieval of relevant studies. Terms such as “quadruped robots”, “legged robotics”, “robot locomotion control”, “robot perception systems”, “robot applications”, “AI in robotics”, and “bio-inspired robotics” were used to query the databases. These keywords ensured a focus on studies addressing core aspects of quadruped robot design and application.

To refine search results further, Boolean operators were applied to construct advanced search queries. The logical expressions used included the following:

$$\begin{array}{cc}\hfill (“quadrupedrobot”\vee “four-leggedrobot”)& \\ \hfill \wedge (“controlmechanisms”\vee “machinelearning”)& \\ \hfill \wedge \neg (“bipedalrobots”).\end{array}$$

This formulation ensured that studies relevant to quadruped robots were included while excluding research exclusively focused on bipedal robots.

Additionally, citation tracking was performed to identify relevant studies cited in major survey papers and seminal works. By examining reference lists of influential publications, additional sources of high relevance were retrieved, ensuring comprehensive coverage of important developments in the field.

Finally, gray literature sources such as patents, technical reports, and industry white papers were considered. These sources provided insights into the latest advancements and commercial implementations of quadruped robotics, supplementing academic research with real-world applications.

2.3. Eligibility Criteria

To maintain the quality and relevance of the study, the following eligibility criteria were applied:

• Inclusion Criteria

• Studies published in peer-reviewed journals, conference proceedings, or reputable industry reports.
• Research works published within the last 20 years (2005–2025), ensuring coverage of recent technological advancements.
• Papers explicitly addressing quadruped robots, including mechanical design, locomotion control, perception systems, and applications.
• Studies discussing artificial intelligence and machine learning applications in quadruped robotics.
• Comparative studies and surveys providing insights into historical and emerging trends in quadruped robotics.

• Exclusion Criteria

• Papers focusing solely on bipedal, hexapod, or wheeled robots, unless they provided relevant insights applicable to quadruped robotics.
• Studies published in languages other than English unless a reliable translation was available.
• Articles with insufficient technical details, such as abstracts or brief conference posters.
• Duplicates or redundant studies presenting similar results without significant new contributions.

2.4. Study Selection

The study selection process was conducted in multiple stages to ensure the inclusion of the most relevant and high-quality research articles. First, an initial screening was performed by reviewing the titles and abstracts of all retrieved studies, allowing for the removal of papers that were clearly irrelevant to the research topic. Following this, a full-text review was conducted on the remaining studies to assess their relevance concerning the established research questions. Studies that met the inclusion criteria proceeded to the quality assessment stage, where they were systematically evaluated based on their research methodology, clarity of presentation, reproducibility, and overall contribution to the field of quadruped robotics. After the quality assessment was completed, the data extraction process was carried out, wherein key details such as study objectives, methodologies employed, primary findings, and conclusions were systematically documented and categorized for further analysis. Finally, the extracted data underwent a synthesis and analysis phase, during which emerging trends, research gaps, and potential future directions in quadruped robotics were identified. By following this structured selection process, our study ensures a comprehensive and systematic exploration of quadruped robots that can offer valuable insights into their mechanical design, control strategies, perception systems, and practical applications.

3. Historical Evolution of Quadruped Robots

Quadruped robots have evolved significantly since their inception, influenced by advancements in robotics, control systems, materials science, and computing power. This section explicitly aims to answer RQ1: How have quadruped robots evolved over time?

3.1. Comparative Perspectives with Other Legged Systems

In the broader context of legged robotics, quadruped robots represent one of several architectural paradigms, each with distinct advantages and tradeoffs. While the recent surge in research and commercialization of quadrupedal systems highlights a promising balance between complexity, mobility, and adaptability, it is important to examine quadruped robots in relation to other legged configurations such as bipeds, hexapods, and multi-legged robots.

Bipedal robots [32] are inspired by human locomotion and consequently offer unmatched potential in environments designed for humans, with applications such as climbing stairs, using tools, and navigating narrow passages. However, they often face significant challenges in terms of dynamic stability and energy efficiency, requiring sophisticated control algorithms and high computational costs in order to maintain balance.

Hexapod robots [33] and other multi-legged systems, on the other hand, benefit from inherent static stability and redundancy. Their ability to maintain support with multiple legs even during a failure or uneven terrain traversal makes them highly robust for applications such as biological research, planetary exploration, and search-and-rescue missions. Nonetheless, their increased mechanical and computational complexity combined lower maneuverability can limit their use in time-sensitive or highly dynamic environments.

Quadruped robots [34] strike a functional compromise between these extremes. They provide enhanced mobility over rough terrains, with better dynamic performance than hexapods and improved stability compared to bipeds. This balance has contributed to their rapid adoption in both academic research and commercial products. While it is difficult to predict long-term dominance, current trends suggest that quadrupeds are not merely a passing phase but rather a foundational platform for future robotic development. Continued advances in actuation, perception, and adaptive control may further solidify their role, although the emergence of hybrid or reconfigurable morphologies could potentially reshape the landscape of legged robotics.

3.2. Early Developments

The origins of quadruped robotics can be traced back to the mid-20th century when researchers began to explore legged locomotion as an alternative to wheeled systems for navigating uneven terrains. One of the earliest examples is the General Electric Walking Truck [35] (see Figure 1a) developed in the 1960s. Although not autonomous—it required direct human control—it demonstrated the potential of legged machines to handle rough terrains where traditional vehicles struggled.

In the early 1980s, interest in legged robots grew with projects such as the Adaptive Suspension Vehicle (ASV) [36] (see Figure 1b) developed at Ohio State University. While the ASV was a six-legged robot, it contributed valuable insights into legged locomotion and control strategies that would influence future quadruped designs. Another notable example from this period is the Mechanical Horse [37] (see Figure 1c), which showcased basic walking mechanisms.

Figure 1 illustrates some of these early quadruped robots.

3.3. Technological Milestones

The 1990s and early 2000s marked significant advancements due to improvements in microprocessors, sensors, and actuators. Researchers began to develop more sophisticated quadruped robots capable of autonomous operation and complex behaviors.

A pivotal milestone was Boston Dynamics’ introduction of BigDog [12] in 2005 (see Figure 2a). Designed for the U.S. military to carry equipment over difficult terrains, BigDog utilized advanced sensors, hydraulic actuators, and onboard computing to navigate and maintain balance autonomously. Its ability to handle rough terrain, recover from disturbances, and carry heavy loads showcased unprecedented mobility and stability for quadruped robots.

Following BigDog’s success, Boston Dynamics continued innovating with models such as LS3 (Legged Squad Support System) and Spot [38] (see Figure 2b). Spot, introduced in 2015, was a smaller, electrically powered quadruped robot designed for commercial applications such as industrial inspection, construction site monitoring, and entertainment. Its commercialization marked a significant shift toward practical deployable quadruped robots in various industries.

Globally, other institutions made notable contributions. The Swiss Federal Institute of Technology (ETH Zurich) developed ANYmal [39] (see Figure 2c), focusing on autonomous navigation and inspection tasks in industrial environments. ANYmal’s modular design and advanced perception capabilities allowed it to operate in challenging conditions such as offshore platforms and underground mines.

The Massachusetts Institute of Technology (MIT) introduced the Cheetah series [40,41] (see Figure 2d), emphasizing high-speed locomotion and dynamic agility. MIT’s Cheetah robots achieved remarkable speeds and could perform complex maneuvers such as jumping over obstacles, highlighting the potential for quadruped robots in dynamic tasks.

3.4. Trends in Design and Application

Over time, quadruped robot designs have evolved to become more efficient, compact, and capable. Early designs were often large and relied on hydraulic systems, limiting their practical applications due to size, weight, and power constraints. Advances in electric actuators and battery technology enabled the development of lighter and more agile robots with extended operational times.

Biomimicry has become a significant trend, with designers drawing inspiration from animal physiology to enhance efficiency and adaptability. This approach has improved joint design, materials selection, and locomotion strategies, allowing robots to better mimic the movements of animals such as dogs, cats, and cheetahs.

Control systems have also advanced, incorporating sophisticated algorithms and artificial intelligence. Machine learning techniques enable robots to adapt to new environments, learn from experiences, and improve performance over time. These capabilities are essential for navigating unpredictable terrains and handling unexpected obstacles.

Quadruped robots’ applications have expanded beyond military and research purposes. They are now utilized in industrial inspection, search and rescue operations, agriculture, and even entertainment. Companies such as Unitree Robotics have commercialized quadruped robots like Laikago (Figure 3a), AlienGo (Figure 3b), Go2 (Figure 3c), and B2 (Figure 3d), making the technology more accessible and fostering innovation in various fields.

The applications of quadruped robots have expanded as technology has advanced.

Table 2. Historical evolution and trends in quadruped robot applications.

Time Period	Primary Applications	Technological Drivers	Notable Examples
1960–1980	Research and Demonstration	Basic mechanical design, foundational locomotion theories	General Electric Walking Truck (1969)
1980	Experimental Prototypes	Advances in control systems, rudimentary sensors	Adaptive Suspension Vehicle (ASV), Ohio State Univ.
2000	Military Logistics	Hydraulic actuators, onboard computing, dynamic balance control	BigDog (2005), Boston Dynamics
2010–2015	Industrial Inspection, Advanced Research	Electric actuators, sensor fusion, AI integration, autonomous navigation	LS3 (2012), Boston Dynamics; ANYmal (2015), ETH Zurich
2016–2018	Commercial Services, Entertainment, Academic Research	Cost reduction, machine learning algorithms, improved batteries	Spot (2016), Boston Dynamics; Laikago (2017), Unitree Robotics; Mini Cheetah (2018), MIT
2019–Present	Public Safety, Construction, Healthcare, Education	Enhanced autonomy, cloud computing, 5G connectivity, human-robot interaction	Spot 2.0 (2019), Boston Dynamics; ANYmal C (2019), ANYbotics; AlienGo (2019), Unitree Robotics; Vision 60 (2020), Ghost Robotics

and Figure 4 summarize the historical evolution and trends in quadruped robot applications, highlighting the primary applications, technological drivers, and notable examples during different time periods.

As shown in Table 2, the evolution of quadruped robots reflects a transition from conceptual prototypes to sophisticated machines capable of complex tasks in real-world environments. The technological drivers have shifted from basic mechanical designs and locomotion theories in the early years to advanced AI integration, enhanced autonomy, and human–robot interaction in more recent times.

3.5. Main of Evolution Trends

The historical progression of quadruped robots demonstrates several key trends:

• Technological Advancements: Improvements in actuators, sensors, computing power, and AI have enabled quadruped robots to become more autonomous, efficient, and capable.
• Application Expansion: Initially focused on research and military applications, quadruped robots now serve in industrial inspection, healthcare, entertainment, and public safety.
• Commercialization and Accessibility: The entry of companies such as Boston Dynamics and Unitree Robotics into the commercial market has made quadruped robots more accessible to businesses and researchers.
• Biomimicry and Design Optimization: Drawing inspiration from animal locomotion has led to more adaptable and efficient robot designs.

These trends highlight the dynamic nature of quadruped robotics and the potential for future growth across various industries.

4. Design Principles and Architecture

Designing quadruped robots requires a multidisciplinary approach that integrates mechanical engineering, control systems, materials science, and electronics, with all of these fields combined to achieve optimal performance across various applications [20]. This section directly addresses RQ2: What are the fundamental principles of design, control mechanisms, and perception systems in quadruped robots?

As illustrated in Figure 5, the proposed subsystem architecture is organized into four hierarchical layers that exchange information and power through bidirectional links:

• High-Level Control and Planning: This layer hosts the global mission logic, consisting of the gait scheduler, path planner, and task planner. It streams abstract commands (desired body velocity, foothold sequence, torque budgets) downwards to the central controller, while status and fault flags are fed back upwards for online replanning.
• Central Controller/CPU: Acting as the “nervous center”, the CPU fuses raw proprioceptive and exteroceptive data, allocates power, and generates low-level joint setpoints. Because the CPU mediates all inter-subsystem traffic, the double-headed arrows emphasize the closed-loop nature of sensing–control–actuation.
• Actuation, Sensing, and Power Subsystems:

  -

  Actuation converts joint commands into forces through motors, hydraulics, or series-elastic actuators (SEAs).

  -

  Sensing aggregates internal feedback (encoders, IMUs, force/torque sensors) with external perception (RGB-D cameras, LiDAR) to estimate state and terrain.

  -

  Power Management regulates energy flow from batteries, fuel cells, or hybrid packs and reports real-time power budgets to the CPU.

  The curved arrows between each of these blocks and the mechanical structure signify that torque, sensor contact forces, and energy are exchanged directly with the robot body, while corresponding measurements return to their respective controllers.
• Mechanical Structure: The integrated chassis and four articulated legs form the physical interface to the environment. Loads, impacts, and terrain irregularities originate here and propagate upward via sensors to influence both actuator commands and high-level planning.

This layered and bidirectionally-coupled design highlights two key principles: (i) perception–control co-design, in which sensing and actuation are inseparable; and (ii) energy–information symbiosis, whereby power availability directly constrains control decisions and motion strategies conversely affect energy draw.

4.1. Mechanical Design

The mechanical design of a quadruped robot is critical to its mobility, stability, and ability to perform tasks. This encompasses leg configurations, actuation mechanisms, and materials. Recent advancements have greatly enhanced the performance and energy efficiency of quadruped robots, particularly in terms of compliant mechanisms and high-efficiency actuators [42,43].

4.1.1. Leg Configurations, Foot Design, and Kinematics

The leg configuration directly influences the robot’s range of motion, stability, and adaptability to different terrains. Common configurations include articulated, compliant, and rigid legs, each with unique mechanical properties and control requirements.

• Articulated Legs: These legs mimic the biological joints of animals, typically consisting of hip, knee, and ankle joints. Articulated legs provide greater flexibility and allow for complex multi-joint movements, improving the robot’s ability to navigate uneven terrain. The MIT Cheetah (see Figure 6a), for instance, employs articulated legs that allow for high-speed running and dynamic maneuvers [44].
• Compliant Legs: These legs integrate elastic elements to absorb shocks and store elastic energy, enhancing energy efficiency and reducing mechanical stress during operation. Robots such as RHex (see Figure 6b) demonstrate the effectiveness of compliant legs in traversing rough terrains, with the ability to minimize impact forces and increase robustness [45].
• Rigid Legs: Early quadruped robots often used rigid legs (see Figure 1), focusing on basic locomotion. While easier to control, rigid legs are less adaptable to uneven surfaces than articulated or compliant legs. However, modern innovations in control algorithms have slightly mitigated these limitations in more recent designs [46].

Table 3. Comparison of main characteristics of articulated, compliant, and rigid leg configurations.

Feature	Articulated Legs	Compliant Legs	Rigid Legs
Structure	Multiple actively controlled joints (hip, knee, ankle)	Integration of elastic elements (springs, tendons)	Rigid links without dedicated compliance
Adaptability to Terrain	High (flexible joint movements)	Moderate (passive adaptation)	Low (requires precise control)
Energy Efficiency	Moderate	High (energy storage and return)	Low
Impact Mitigation	Active control strategies needed	Passive shock absorption	Limited, requires fast feedback control
Control Complexity	High (multi-DoF coordination)	Moderate (reduced control needs)	Low (simpler kinematic control)
Mechanical Complexity	High	Moderate	Low
Typical Examples	MIT Cheetah [44]	RHex [45]	Early quadrupeds [46]

compares the main characteristics of these three types of leg configurations, highlighting their respective advantages and challenges in terms of structure, adaptability, energy efficiency, impact mitigation, and control complexity.

The choice between articulated, compliant, or quasi-rigid legs in quadrupedal robots is fundamentally driven by several key considerations: (i) workspace requirements such as stride length s and ground clearance $h_{\mathrm{clr}}$; (ii) actuator placement and gearing strategies; (iii) intrinsic versus extrinsic compliance; and (iv) the adopted impact mitigation strategy. Each design exhibits distinct strengths; articulated legs typically offer greater adaptability and precise trajectory control due to multiple actively controlled degrees of freedom [44], compliant legs enhance energy efficiency and passive shock absorption [47], and quasi-rigid legs simplify control but limit adaptability [39].

An important ongoing research direction involves integrating the advantages of these distinct leg configurations through hybrid designs. Hybrid leg designs, combining active joint articulation with passive compliance mechanisms, aim to achieve a balance among mechanical complexity, energy efficiency, and terrain adaptability [48,49]. For instance, prototypes integrating articulated hip joints with compliant knee or ankle segments have demonstrated promising results, achieving both agility and robustness [39,47]. Recent studies have shown that incorporating parallel or series elastic actuators into articulated structures significantly enhances locomotion efficiency and impact resilience [39,49]. The combination of active actuation for precise trajectory tracking and passive elasticity for energy storage and release leads to optimized dynamic performance, particularly beneficial in uneven and unpredictable terrains [50,51]. Realizing the full potential of these hybrid configurations requires advanced control methods.

Beyond the general leg structure, foot design plays a crucial role in the robot’s ability to interact with the environment. Advanced footpads enhance traction, absorb impacts, and adapt to irregular terrain features. Designs range from flat rubber soles for urban applications to clawed or toe-like structures for soft or granular surfaces. Some robots, such as ANYmal, integrate force sensors or tactile elements in the feet to support real-time gait adaptation and terrain classification [52]. Additionally, there is growing interest in hybrid leg–wheel mechanisms that allow robots to switch between walking and rolling modes. These configurations aim to combine the terrain adaptability of legs with the energy efficiency and speed of wheels. Although not yet mainstream in quadrupedal platforms, early demonstrations such as RoboSimian [53] highlight the potential of such systems for diverse operational scenarios, including urban exploration and industrial inspection.

Kinematic analysis is fundamental to designing leg movements, ensuring that the robot can maintain balance and follow precise trajectories. The forward and inverse kinematics equations enable the calculation of foot positions and joint angles, which are critical for effective motion planning [40].

4.1.2. Actuation Systems

Actuators are responsible for joint movement, significantly influencing the robot’s performance characteristics. The main types of actuators used in quadruped robots include:

• Electric Motors: In particular, brushless DC motors are widely used for their precision and controllability. Robots such as ANYmal use electric actuators to achieve fine motion control while maintaining a compact design [52]. These actuators are energy-efficient and well-suited for indoor and research environments where noise and precision are key considerations.
• Hydraulic Actuators: These are known for their high power density, and are typically used in larger quadruped robots that require significant force output. However, hydraulic systems are generally heavier and less energy-efficient. Boston Dynamics’ BigDog, for example, employs hydraulic actuators to achieve robust performance on rough terrain despite the tradeoffs in weight and complexity [12].
• Series Elastic Actuators (SEAs): These incorporate an elastic element between the motor and load, which improves force control and shock tolerance. They are commonly used in dynamic robots such as the MIT Cheetah for better energy efficiency and impact handling, allowing the robot to absorb shocks and conserve energy during gait cycles [44].

The choice of actuator system directly affects the robot’s speed, agility, and energy consumption. Innovations in actuator technology such as high-torque-density electric motors and lightweight hydraulic systems have enabled quadruped robots to achieve unprecedented levels of agility and efficiency [41].

4.1.3. Materials and Structural Design

Material selection and structural design are critical for reducing the robot’s weight while maintaining strength and durability. Common materials include:

• Lightweight Metals: Aluminum and titanium alloys are widely used due to their high strength-to-weight ratios. Aluminum is preferred for structural components in environments where cost and ease of manufacturing are important, while titanium is more suitable for extreme conditions due to its superior strength and corrosion resistance [54,55].
• Composites: Carbon fiber composites are frequently used in high-performance robots such as the MIT Cheetah, providing an excellent strength-to-weight ratio. Although more expensive, composites allow for complex shapes and reduce the robot’s overall weight, enhancing its dynamic capabilities [34].
• Polymers: High-strength polymers are commonly used for non-load-bearing components or protective enclosures. Advanced polymers, e.g., reinforced nylon, can balance weight and mechanical properties for specific structural elements [56].

Structural optimization techniques such as topology optimization are increasingly being used to design lightweight and efficient structures tailored to specific load conditions [57]. Additive manufacturing (3D printing) also plays a crucial role, enabling engineers to fabricate complex geometries that optimize the robot’s structural integrity while minimizing its weight. For instance, lattice structures [58] provide strength where needed while reducing mass, which is crucial for dynamic performance.

The combination of lightweight metals, composites, and polymers with advanced manufacturing techniques enables the development of agile and robust quadruped robots capable of operating in diverse environments. The ANYmal robot, for example, combines aluminum alloys and carbon fiber composites to optimize weight and strength, achieving efficient locomotion while carrying payloads [52]. Integrating these materials with bio-inspired structural designs such as flexible spines or articulated joints modeled after animal anatomy can significantly enhance the performance and adaptability of quadruped robots [46].

4.2. Sensory Systems

Sensory systems (see Figure 7 and Table 4) allow quadruped robots to perceive both their environment and their own state, which is essential for autonomous operation and interaction with surroundings. Effective sensing enables robots to navigate complex terrains, avoid obstacles, maintain balance, and perform tasks that require interaction with objects or humans.

4.2.1. Internal Sensors (Proprioception)

Internal sensors provide feedback on the robot’s own movements and positions, which is crucial for precise control and stability. The most common internal sensors are reported below:

• Incremental Rotary Encoders: Encoders [59] measure joint angles and rotational speeds, enabling precise motion control. They are typically installed on each joint actuator to provide real-time feedback on the position and velocity of joints. High-resolution encoders improve the accuracy of kinematic calculations and are essential for tasks requiring fine manipulation or precise foot placement.
• Force/Torque Sensors: These sensors [60] detect forces and torques at joints or contact points, aiding in balance and interaction with objects. By measuring the interaction forces between the robot and the environment, force/torque sensors enable compliance control, which allows the robot to adjust its movements in response to external forces. This is particularly important for tasks involving physical contact, such as climbing stairs or handling objects.
• Inertial Measurement Units (IMUs): IMUs [61] provide data on orientation, acceleration, and angular velocity, crucial for maintaining balance and posture. They typically combine accelerometers, gyroscopes, and sometimes magnetometers to estimate the robot’s attitude and motion. The data from IMUs is used in state estimation algorithms to correct for drift and improve the accuracy of the robot’s pose estimation.

Proprioceptive sensing is vital for implementing control algorithms that ensure stability and smooth locomotion. For example, feedback from encoders and IMUs is used in closed-loop control systems to adjust joint commands in real-time, helping the robot to maintain balance even when subjected to disturbances. Additionally, internal sensors enable the implementation of sophisticated locomotion algorithms such as MPC and whole-body control, which require accurate knowledge of the robot’s state.

4.2.2. External Sensors (Exteroception)

External sensors gather information about the environment, allowing the robot to perceive obstacles, navigate, and interact with objects.

• LiDAR (Light Detection and Ranging): LiDAR sensors [62] generate precise 3D maps of the environment by emitting laser pulses and measuring the time required for the reflections to return. This technology enables high-resolution spatial awareness, which is essential for navigation and obstacle avoidance in complex terrains. LiDAR is particularly useful in low-light conditions where cameras may be less effective.
• Cameras: Cameras [5,6,7] capture visual information that is useful for object recognition, terrain assessment, and Simultaneous Localization and Mapping (SLAM). Stereo cameras provide depth information by comparing images from two slightly offset lenses, while monocular cameras rely on visual cues and algorithms to estimate depth. RGB-D cameras combine traditional RGB data with depth information, providing more comprehensive environmental understanding. Advanced computer vision techniques, including Convolutional Neural Networks (CNNs), are employed to process camera data for tasks such as semantic segmentation and object detection.
• Ultrasonic Sensors: Ultrasonic sensors [63] measure distances to nearby objects using sound waves. They are cost-effective and useful for short-range obstacle detection. However, their resolution and accuracy are lower compared to LiDAR and cameras. Ultrasonic sensors are often used as supplementary sensors to provide redundancy and enhance safety.
• Tactile Sensors: Tactile sensors [64] detect contact forces and textures, enhancing interaction capabilities. These sensors can be placed on the robot’s feet or manipulator to provide feedback during physical interactions. Tactile sensing enables the robot to adjust its grip on objects, detect slippage, and perform delicate tasks that require force control.
• GPS (Global Positioning System): For outdoor navigation, GPS [65] provides global position information. While GPS lacks the precision needed for fine control, it is useful for high-level navigation tasks, such as moving to a specific location in open environments.
• Infrared Sensors: Infrared (IR) sensors [66] can detect heat signatures and are useful for applications such as search and rescue, where locating warm bodies is critical.

Table 4. Summary of key sensors used on quadruped robots (reorganized).

Sensor Type	Advantages	Limitations	Typical Use Cases
Internal Sensors (Proprioception)
Incremental Rotary Encoders [59]	Real-time joint position and velocity feedbackHigh resolution for precise motion control	Measure only rotational motion (no external feedback)Require careful calibration and installation	Joint angle measurementFine manipulation or precise foot placement
Force-Torque Sensors [60]	Measure contact forces and joint torquesEnable compliance control for safe interactions	Often expensive; prone to noise or drift if not calibratedInstallation can be mechanically complex	Balance and disturbance rejectionStair climbing, object handlingAny task requiring force feedback
Inertial Measurement Units (IMUs) [61]	Provide orientation, acceleration, and angular velocityCompact and reliable	Accumulative drift over timeNo direct perception of external environment	Robot posture control and stabilizationState estimation (sensor fusion)
External Sensors (Exteroception)
LiDAR (Light Detection and Ranging) [62]	High-resolution 3D mappingOperates well in low-light conditionsPrecise distance measurements	High costPerformance degradation in rain, fog, or reflective surfaces	Autonomous navigation, obstacle avoidance3D mapping in industrial or outdoor settings
Cameras (Monocular, Stereo, RGB-D) [5,6,7,67,68,69,70,71,72,73]	Rich visual informationEnables object recognition and SLAMRGB-D offers combined color + depth data	Monocular lacks direct depthSensitive to lighting changes and occlusionsRGB-D sensors often have limited range	Visual SLAM, semantic segmentation, object detectionTerrain assessment and obstacle avoidance
Ultrasonic Sensors [63]	Low costEffective for short-range obstacle detection	Limited resolutionUnsuitable for detailed mapping or complex environments	Proximity sensingRedundant safety checks
Tactile Sensors [64]	Detect contact forces, textures, and slippageEnhance manipulation capabilities	Typically short-range or surface-level feedback onlyIntegration can be mechanically challenging	Precise object handlingSlip detectionForce-based exploration
GPS (Global Positioning System) [65]	Provides global position in open outdoor areasUseful for high-level navigation	Ineffective indoors or with obstructed satellitesLimited precision for fine control	Outdoor path planning, waypoint navigationLarge-scale field operations
Infrared (Thermal) Sensors [66]	Capable of detecting heat signaturesUseful in low-visibility scenarios	Sensitive to thermal noise or reflective surfacesLimited structural information	Search and rescue (locating warm bodies)Nighttime or smoke-filled environment navigation

Table 4. Summary of key sensors used on quadruped robots (reorganized).

Sensor Type	Advantages	Limitations	Typical Use Cases
Internal Sensors (Proprioception)
Incremental Rotary Encoders [59]	Real-time joint position and velocity feedbackHigh resolution for precise motion control	Measure only rotational motion (no external feedback)Require careful calibration and installation	Joint angle measurementFine manipulation or precise foot placement
Force-Torque Sensors [60]	Measure contact forces and joint torquesEnable compliance control for safe interactions	Often expensive; prone to noise or drift if not calibratedInstallation can be mechanically complex	Balance and disturbance rejectionStair climbing, object handlingAny task requiring force feedback
Inertial Measurement Units (IMUs) [61]	Provide orientation, acceleration, and angular velocityCompact and reliable	Accumulative drift over timeNo direct perception of external environment	Robot posture control and stabilizationState estimation (sensor fusion)
External Sensors (Exteroception)
LiDAR (Light Detection and Ranging) [62]	High-resolution 3D mappingOperates well in low-light conditionsPrecise distance measurements	High costPerformance degradation in rain, fog, or reflective surfaces	Autonomous navigation, obstacle avoidance3D mapping in industrial or outdoor settings
Cameras (Monocular, Stereo, RGB-D) [5,6,7,67,68,69,70,71,72,73]	Rich visual informationEnables object recognition and SLAMRGB-D offers combined color + depth data	Monocular lacks direct depthSensitive to lighting changes and occlusionsRGB-D sensors often have limited range	Visual SLAM, semantic segmentation, object detectionTerrain assessment and obstacle avoidance
Ultrasonic Sensors [63]	Low costEffective for short-range obstacle detection	Limited resolutionUnsuitable for detailed mapping or complex environments	Proximity sensingRedundant safety checks
Tactile Sensors [64]	Detect contact forces, textures, and slippageEnhance manipulation capabilities	Typically short-range or surface-level feedback onlyIntegration can be mechanically challenging	Precise object handlingSlip detectionForce-based exploration
GPS (Global Positioning System) [65]	Provides global position in open outdoor areasUseful for high-level navigation	Ineffective indoors or with obstructed satellitesLimited precision for fine control	Outdoor path planning, waypoint navigationLarge-scale field operations
Infrared (Thermal) Sensors [66]	Capable of detecting heat signaturesUseful in low-visibility scenarios	Sensitive to thermal noise or reflective surfacesLimited structural information	Search and rescue (locating warm bodies)Nighttime or smoke-filled environment navigation

Integrating multiple sensor modalities enhances environmental perception, enabling complex tasks such as SLAM and advanced obstacle avoidance. Sensor fusion techniques combine data from various sensors to improve accuracy and robustness. For example, combining LiDAR and camera data [72] can provide both precise distance measurements and rich visual information, allowing the robot to navigate while recognizing and categorizing objects in its environment.

As an instance, in the ANYmal robot [52], a combination of LiDAR, cameras, IMUs, and joint encoders is used to achieve autonomous navigation and inspection tasks. The LiDAR provides detailed 3D mapping for obstacle avoidance, while cameras are used for visual inspection and object recognition. The integration of these sensors allows ANYmal to navigate complex industrial environments and perform tasks such as reading gauges or detecting anomalies.

4.3. Power and Energy Management

Efficient power systems are essential for the autonomy and operational endurance of quadruped robots. Energy management influences the robot’s operational time, mobility, and the feasibility of performing tasks in different environments.

4.3.1. Power Sources

The main power sources (see

Table 5. Comparison of power sources for quadruped robots.

Power Source	Advantages	Limitations	Typical Use Cases
Lithium-Ion Battery [38,74,75]	High energy density and lightweight Rechargeable and widely available	Limited operational time (1–3 h) Sensitive to temperature variations	Suitable for lightweight robots used in indoor and short-term outdoor applications, e.g., inspection robots like Spot.
Fuel Cell [76,77,78]	Extremely high energy density (800–1000 Wh/kg) Long operational time (8–12 h)	High cost and complex maintenance Requires hydrogen storage and refueling infrastructure	Ideal for long-duration missions in remote areas, e.g., exploration or disaster response tasks.
Hybrid System [79,80,81]	Combines benefits of batteries and fuel cells Provides steady energy with bursts of high power	Increased weight due to dual power systems Higher system complexity	Useful for tasks requiring dynamic performance and longer operational time, e.g., search and rescue operations.

) used for quadruped robots are listed below.

• Batteries: Lithium-ion batteries [74,75] are commonly used due to their high energy density and rechargeability. They are suitable for robots requiring moderate power for extended periods. Battery technology advancements have led to lighter and more efficient energy storage solutions. For example, the Boston Dynamics Spot [38] robot uses rechargeable lithium-ion batteries, providing about 90 min of operation per charge.
• Fuel Cells: Fuel cells [76,77,78] offer higher energy densities and longer operation times but are less common due to cost and complexity. They generate electricity through chemical reactions, typically using hydrogen. Fuel cells [78] are suitable for applications requiring long-duration missions without the possibility of recharging, such as remote exploration.
• Hybrid Systems: Hybrid power systems [79,80,81] combine different power sources such as batteries and supercapacitors to handle varying power demands. Supercapacitors can deliver high-power bursts for dynamic movements, while batteries provide steady energy for regular operation. Hybrid systems aim to optimize energy efficiency and extend operational time.

The choice of power source impacts the robot’s weight, size, operational time, and suitability for specific tasks. Trade-offs between energy density, power output, weight, and recharging/refueling logistics must be considered during design.

4.3.2. Energy Efficiency Strategies

To extend operational time and improve efficiency, various strategies are employed:

• Regenerative Braking: This technique [82] recovers energy during the deceleration phases. When the robot’s joints slow down or stop moving, the kinetic energy is converted back into electrical energy and stored into the battery. Regenerative systems [83] are more common in electrically actuated robots.
• Energy-Efficient Gaits: Optimizing locomotion patterns reduces energy consumption [84,85]. Researchers can study animal locomotion to develop gaits that minimize energy expenditure [86]. For instance, using passive dynamics and exploiting the robot’s natural dynamics can lead to more efficient movement [87,88].
• Adaptive Control Systems: These systems adjust power usage based on task requirements and environmental conditions [89]. For example, reducing movement speed or disabling non-essential sensors [84] when battery levels are low. Intelligent power management algorithms balance performance with energy consumption.
• Lightweight Design: Reducing the robot’s weight through material selection and structural optimization decreases the energy required for movement [55]. As discussed in the mechanical design section, lighter robots consume less power for the same tasks.
• Advanced Actuators: Using high-efficiency actuators [90] such as optimized electric motors or series of elastic actuators can reduce power consumption. Actuators that are able to store and release elastic energy contribute to overall energy efficiency [91].

4.4. Integration of Systems

Designing a quadruped robot requires the seamless integration of mechanical components, sensors, power systems, and control algorithms. This integration ensures that all subsystems work harmoniously to achieve the desired performance, reliability, and functionality. Key considerations include:

• Modularity: Designing components that can be easily replaced or upgraded enhances maintainability and adaptability [92,93]. Modular design allows for rapid prototyping and testing of different subsystems without affecting the entire system. For example, the ANYmal robot [39] utilizes a modular approach in which each leg can be detached and serviced independently, reducing downtime and facilitating upgrades.
• Weight Distribution: Proper weight balance is crucial for stability and efficient locomotion. The distribution of mass affects the robot’s center of gravity, which in turn influences balance and agility [94,95]. Engineers must carefully design the placement of heavy components such as batteries and actuators to optimize performance, and users should also pay attention to the placement of payloads. Computational tools and simulations are often used to analyze the effects of weight distribution on dynamic behavior.
• Communication Networks: Reliable and fast data exchange between sensors, actuators, and control units is essential for real-time operation [96]. Communication protocols must support high data rates with minimal latency to ensure timely processing of sensor inputs and execution of control commands. Common communication interfaces include CAN bus, ethernet, and wireless networks for remote operation. The integration of distributed control systems requires robust synchronization and error-handling mechanisms [97].
• Software Architecture: A well-designed software framework is critical for integrating various algorithms, from low-level motor control to high-level decision-making. Middleware platforms such as the Robot Operating System (ROS) [98,99] provide standardized interfaces and tools for developing modular and reusable software components. This facilitates collaboration and scalability, allowing developers to focus on specific functionalities without redefining common processes.
• Power Management: Coordinating power distribution among different subsystems is vital to prevent overloads and ensure efficiency. Power management systems monitor energy consumption, control voltage levels, and protect against faults. Intelligent power allocation can prioritize critical functions during low-power conditions, enhancing the robot’s operational resilience.
• Thermal Management: Heat generated by actuators, processors, and other electronics must be effectively dissipated to maintain optimal performance. Thermal management strategies include heat sinks, fans, and thermal interface materials. In some designs, the robot’s structure is used to conduct heat away from sensitive components [100].
• Safety Systems: Integration of safety features such as emergency stop mechanisms [101], collision detection, and compliance control is essential for operating in environments with humans or delicate equipment. Redundant systems and failsafes help to prevent accidents resulting from component failures or unexpected conditions.

5. Perception, Sensing, and Environment Interaction

Quadruped robots rely heavily on advanced perception and sensing systems to interact effectively with their environment. These systems enable robots to navigate complex terrains, avoid obstacles, and perform tasks autonomously. Recent advancements in sensor technology and computational algorithms have significantly enhanced the capabilities of quadruped robots in perceiving and understanding their surroundings [12]. Figure 8 illustrates the role of perception and environmental interaction in enhancing quadruped robots’ capabilities through advanced sensing technologies. This section contributes to RQ2 and RQ3: How do quadruped robots integrate perception and AI-based adaptability?

Table 6. Summary of key algorithms for perception, sensing, and environment interaction in quadruped robots.

Category	Algorithms and References	Advantages	Applications
SLAM (Simultaneous Localization and Mapping)	LiDAR-based SLAM (e.g., LOAM) [102], Visual SLAM (e.g., ORB-SLAM) [103], RGB-D SLAM [104], Graph-based SLAM [105], Learning-based SLAM [106]	Real-time mapping and localization, robust in GPS-denied environments, high accuracy in diverse terrains	Autonomous navigation, exploration of unknown or dynamic environments, search and rescue operations
Terrain Perception and Classification	Depth-based footstep planning [107], proprioceptive-based classification (RNN, CNN) [108], audio-based terrain recognition [109]	Adapts gait in real-time, improves stability on varied surfaces, reduces need for manual feature engineering	Off-road exploration, agricultural applications, long-distance field autonomy
Object Recognition and Interaction	CNN-based detection (e.g., Faster R-CNN, YOLO) [110], RGB-D object recognition [111], semantic segmentation (FCN, SegNet) [112,113]	High accuracy in object detection, depth cues improve pose estimation, enables complex manipulation tasks	Pick-and-place, inspection and maintenance, human-robot collaboration
Obstacle Detection and Avoidance	Sensor fusion (LiDAR + vision) [114,115], ultrasonic-based detection [116], depth camera-based avoidance [111]	Multi-modal data for robust detection, real-time collision avoidance, suitable for unstructured or dynamic settings	Indoor/outdoor navigation, disaster response, warehouse logistics
Path Planning	A* [117], RRT/RRT* [118,119], Dynamic Window Approach (DWA) [120], Probabilistic Roadmaps (PRM) [121]	Efficient route computation, handles kinodynamic constraints, adaptable to real-time changes	Complex terrain navigation, urban search and rescue, autonomous exploration
Human-Robot Interaction (HRI)	Communication interfaces (visual, auditory, tactile) [122,123], safety algorithms (collision avoidance) [124], emotion recognition and expression [125]	Enhances user trust and acceptance, facilitates intuitive operation, ensures safety and comfort	Assistive robotics, collaborative tasks in shared spaces, search and rescue with remote teleoperation
Machine Learning for Perception and Decision-Making	RL [126,127], supervised learning (CNN, RNN) [108,128], unsupervised learning (Autoencoder, GAN) [129,130]	Learns adaptive and robust policies, reduces reliance on hand-engineered features, generalizes to new tasks or terrains	Gait optimization, terrain classification, complex manipulation

shows the various sensing and recognition algorithms used in quadruped robots and how they contribute to environmental interaction.

5.1. Environmental Mapping and Sensing

Accurate environmental mapping is essential for quadruped robots to understand and navigate their surroundings using different types of sensors (see Table 4). By creating detailed representations of the environment, robots can plan paths, avoid obstacles, and adapt their locomotion strategies. Several key technologies and methods are outlined below.

5.1.1. Simultaneous Localization and Mapping (SLAM)

SLAM algorithms [131,132,133] enable robots to simultaneously build a map of an unknown environment while tracking their location within it, which is crucial for autonomous navigation in GPS-deprived environments. Several types of SLAM approaches are commonly applied in robotics.

LiDAR-based SLAM relies on LiDAR sensors to create detailed 3D maps. LiDAR provides high-resolution depth information, making it particularly useful in environments with poor lighting or textureless surfaces, where cameras may face challenges [102]. For instance, Zhang and Singh developed Lidar Odometry and Mapping (LOAM), a real-time method that has been widely adopted in robotics [102].

In contrast, Visual SLAM uses cameras to extract visual features from images and track their movement over time. Visual SLAM tends to be more lightweight and cost-effective, but is sensitive to lighting conditions and requires robust feature detection algorithms [103]. ORB-SLAM is a notable example that offers accurate monocular SLAM through fast feature extraction [103].

RGB-D SLAM combines color and depth information from RGB-D cameras such as the Microsoft Kinect to improve mapping accuracy. By leveraging both visual and depth cues, this approach enhances performance in complex environments [104]. Endres et al. developed an RGB-D SLAM system that integrates depth data to generate dense 3D maps [104].

In addition, advanced SLAM algorithms incorporate probabilistic methods and machine learning techniques to improve robustness and adaptability. For example, graph-based SLAM optimizes the robot’s trajectory and map by minimizing errors across multiple sensor measurements [105]. Deep learning techniques have also been integrated into SLAM to enhance feature extraction and data association [106].

5.1.2. Terrain Perception and Classification

Understanding the type and condition of terrain is crucial for adapting locomotion strategies and ensuring stability. Quadruped robots must perceive and classify terrain in real-time in order to adjust their gait and maintain balance [134]. Depth sensors provide key information about the topography ahead, allowing the robot to plan foot placements effectively. Stereoscopic cameras or structured light sensors can detect variations in ground height, which is essential for adjusting movements. For instance, Wellhausen et al. [107] demonstrated terrain-dependent locomotion by integrating depth perception for footstep planning, showing the importance of terrain awareness in navigation.

In addition to depth perception, terrain classification algorithms analyze sensor data to categorize terrain types such as flat, sloped, rough, or slippery surfaces, enabling adaptive gait selection. Machine learning techniques are used to classify terrain based on visual and tactile inputs [108]. Recent research by Vulpi et al. [108] has highlighted that field robots can improve their long-distance autonomy by accurately classifying terrain using only proprioceptive signals and deep neural networks such as Recurrent Neural Networks (RNNs) or CNNs, without the need for visual data. This approach outperforms traditional Support Vector Machines (SVMs) by eliminating the requirement for predefined feature spaces [108]. Additionally, Valada et al. [109] proposed a deep Long Short-Term Memory (LSTM) recurrent model that leverages vehicle–terrain interaction sounds to capture spatial and temporal dynamics. This method improves both the accuracy and robustness of autonomous robot terrain classification even in high-noise environments, overcoming the limitations of previous methods.

Force and torque feedback from sensors embedded in the robot’s legs also play an essential role in terrain perception. These sensors detect ground contact and compliance, informing necessary adjustments in balance and posture. This proprioceptive feedback allows the robot to respond to unexpected changes in terrain such as loose rocks or soft soil [135]. For example, the ANYmal robot utilizes torque-controlled actuators to sense and adapt to ground interactions in real time [39].

Integrating exteroceptive sensors such as vision and LiDAR with proprioceptive feedback like force and joint position sensing further enhances a robot’s ability to navigate challenging environments. Recent advancements have focused on sensor fusion techniques, combining multiple sources of information for more robust terrain perception and improved navigation capabilities [136].

5.2. Object Recognition and Interaction

Quadruped robots often need to recognize and interact with objects within their environment, especially in tasks such as manipulation, inspection, search and rescue, or human–robot interaction. Advanced perception and manipulation capabilities enable these robots to perform complex tasks in unstructured environments [31,137]. This section discusses the key technologies that facilitate object recognition and interaction in quadruped robots.

5.2.1. Visual Recognition Systems

Effective object recognition is essential for quadruped robots to understand their surroundings and make informed decisions. One of the key technologies in this domain is the use of deep learning neural networks. In particular, CNNs are widely employed for image classification and object detection tasks, allowing robots to identify and locate objects within camera images. This capability is crucial for tasks such as picking up objects or navigating around obstacles [8,9,10]. For example, CNN-based algorithms like Faster R-CNN [138,139,140] enable real-time object detection on robotic platforms, making them highly effective in dynamic environments.

Depth cameras such as RGB-D sensors are another critical component of visual recognition systems. These cameras provide spatial information that enhances the accuracy of object recognition by combining depth data with visual information. This combination allows for better estimation of object size, position, and orientation, which is especially important for grasping and manipulation tasks [111,141,142]. By leveraging this fusion of data, robots can improve their ability to interact with objects in their environment [143,144].

Another important technology is semantic segmentation, which assigns a class label to each pixel in an image. This enables the robot to distinguish between different objects and surfaces in its environment, which is a critical capability that for tasks such as obstacle avoidance and path planning [112,145]. Fully Convolutional Networks (FCNs) have been developed to facilitate accurate segmentation, further enhancing robots’ ability to navigate complex environments [113,146].

By integrating different visual recognition systems—deep learning for object detection, depth cameras for spatial information, and semantic segmentation for object differentiation—quadruped robots can develop a comprehensive understanding of their surroundings, resulting in greater autonomy and more effective interaction with their environment.

5.2.2. Manipulation Capabilities

Some quadruped robots are equipped with manipulators or grippers, allowing them to interact with objects and extending their functionality beyond locomotion. One of the key features enabling this interaction is the use of arm attachments. Robotic arms integrated into quadruped platforms allow robots to perform tasks such as opening doors, pressing buttons, or picking up and carrying items [43,147]. The design of these manipulators must carefully account for factors such as payload capacity, degrees of freedom, and control complexity in order to ensure smooth coordination between the arm and the robot’s locomotion system [23,136,148].

Compliance control is another essential feature that ensures safe interaction with both objects and humans. By adjusting the force and motion of the manipulator, compliance control strategies enable the robot to handle delicate objects without causing damage, while also adapting to unstructured environments [149,150,151]. This capability is especially important in scenarios involving human–robot collaboration, where safety and adaptability are critical.

In addition to physical manipulation, hand–eye coordination plays a crucial role in tasks that require precise positioning, such as assembling components or using tools. This coordination involves integrating vision systems with the robot’s manipulators, allowing for synchronized movements based on visual feedback [152,153]. Hand–eye coordination algorithms help to enhance robots’ accuracy and efficiency during these tasks, significantly improving their ability to perform complex manipulations [154].

Advancements in manipulation capabilities greatly increase the versatility of quadruped robots, enabling them to perform a wider range of complex tasks that require both mobility and dexterity.

5.3. Obstacle Detection and Avoidance

In order to operate safely and efficiently, quadruped robots must detect and navigate around obstacles in dynamic and unstructured environments. Effective obstacle detection and avoidance are crucial for maintaining stability, preventing collisions, and ensuring mission success [155]. This section explores the sensor modalities and path-planning algorithms that enable quadruped robots to perceive obstacles and plan safe trajectories.

5.3.1. Sensor Modalities for Obstacle Detection

Obstacle detection is a critical capability for quadruped robots that enables safe navigation in complex and dynamic environments. Section 4.2.2 provides a detailed description of the various sensor modalities used to achieve robust obstacle perception and interaction. This section summarizes and highlights their specific applications in obstacle detection and avoidance.

For example, LiDAR sensors provide high-resolution distance measurements by emitting laser pulses and calculating the time it takes for reflections to return, generating precise 3D maps of the surroundings [102]. LiDAR is can capture detailed environmental features and is particularly effective in varied lighting conditions, making it well-suited for navigating complex terrains [156].

Ultrasonic sensors, on the other hand, detect obstacles by emitting high-frequency sound waves and measuring the echo time. These sensors are cost-effective and useful for short-range obstacle detection [116]. However, they have lower resolution than LiDAR and can be influenced by environmental noise, limiting their effectiveness in more complex or noisy settings.

Cameras are another important sensor modality for obstacle detection. Visual cameras capture images that can be processed using machine learning algorithms to detect and recognize obstacles such as objects, pedestrians, and other robots [110]. While cameras enable rich visual analysis, their performance can be affected by factors such as lighting conditions and occlusions, which may hinder their reliability in certain environments. Depth cameras such as stereo cameras or RGB-D sensors enhance obstacle detection by providing depth information alongside visual data [111,141,142]. The addition of depth information allows for more accurate localization and size estimation of obstacles, improving the robot’s ability to navigate around them.

Integrating multiple sensor modalities through sensor fusion techniques significantly improves the robustness and reliability of obstacle detection [114]. By combining the strengths of different sensors, sensor fusion compensates for the limitations of individual modalities and allows for more comprehensive and accurate obstacle detection [115].

5.3.2. Path Planning Algorithms

When obstacles are detected, path planning algorithms are responsible for computing optimal paths while avoiding collisions. One of the most commonly used algorithms is the A* algorithm, a heuristic search method that finds the shortest path in a grid-based map by minimizing the cost from the start node to the goal node [117]. A* is widely favored for its simplicity and effectiveness in static environments, although its performance largely depends on the heuristic function used to estimate the cost to the goal [157].

In more complex and dynamic environments, Rapidly-exploring Random Trees (RRT) is often employed. RRT is an algorithm that efficiently explores high-dimensional spaces by randomly building a space-filling tree [118]. This makes it particularly suitable for quadruped robots navigating through cluttered or unknown environments, as it can handle kinodynamic constraints [119,158].

For real-time obstacle avoidance, the Dynamic Window Approach (DWA) is a commonly used algorithm. DWA considers the robot’s kinematic constraints to generate feasible motion commands in real time [120]. It evaluates possible velocities within the robot’s dynamic limits, helping the robot to avoid obstacles while moving toward its goal [159,160,161]. This makes it particularly effective in dynamic environments where real-time decision-making is crucial.

Another approach is the Probabilistic Roadmap (PRM), which builds a graph of possible paths by randomly sampling the configuration space and connecting feasible paths [121]. PRMs are particularly useful in high-dimensional spaces, and are often employed for offline planning in known environments.

In advanced path planning, a combination of global and local planning techniques is often used. Global planning algorithms such as A* and PRM are typically used for long-term navigation, while local planning methods like DWA are employed for immediate obstacle avoidance [162,163]. Additionally, machine learning techniques are increasingly being integrated into path planning algorithms to improve adaptability and decision-making in dynamic environments [164].

5.4. Human-Robot Interaction

Interaction with humans is crucial for quadruped robots operating in shared environments. Effective communication and safety are paramount to ensure seamless collaboration and acceptance by human users [165]. Research into Human–Robot Interaction (HRI) focuses on developing systems that facilitate intuitive and safe interactions between robots and humans [166].

5.4.1. Communication Interfaces

Effective communication interfaces enable quadruped robots to convey their intentions, status, and responses to human collaborators, enhancing transparency and trust in human–robot interactions [167]. These interfaces make interactions more natural and efficient, allowing humans to better understand and predict the robot’s actions.

One common communication method is through visual signals. Robots can use LEDs, screens, or robotic gestures to convey information [168]. For instance, changing light colors can indicate the robot’s operational status, while displays can show specific messages to inform users of the robot’s current state. Additionally, robotic gestures can non-verbally communicate intentions or responses, making interactions feel more intuitive [169].

Auditory signals such as voice output or alerts are another effective modality. By implementing speakers, robots can provide verbal feedback, warnings, or instructions [170]. These auditory cues are particularly useful in situations where visual attention is limited, helping to ensure that important information reaches human collaborators and fostering a sense of trust through clear communication.

For remote operation, control interfaces such as tablets, smartphones, and specialized controllers are commonly used [122]. These user-friendly interfaces allow operators to control the robot intuitively, often incorporating live video feeds and touch-based controls. This is particularly important in mission-critical applications such as search and rescue, where precise and responsive control is essential [171].

Natural Language Processing (NLP) also plays a vital role in communication interfaces. With NLP, robots can understand and respond to spoken commands, making interactions more natural and reducing the learning curve for users [172]. This technology allows for complex instructions to be conveyed efficiently, making it easier for non-expert users to work with robots [173].

Tactile interfaces are another emerging area of HRI, focusing on physical contact between humans and robots. Tactile sensing enables robots to detect and respond to various forms of human touch, such as guidance, instruction, or emotional expression [123]. Research into tactile HRI categorizes contact interactions into three main types: those that interfere with robot behavior, those that contribute to behavior execution, and those that facilitate behavior development [123]. By integrating tactile sensors such as force/torque sensors, resistive arrays, and capacitive sensors, quadruped robots can perceive touch and respond appropriately, enhancing their capacity for safe and natural interactions [123].

Emotional intelligence is increasingly being incorporated into robot design. This involves integrating social cues and behaviors that make interactions with robots more comfortable and natural. Robots that can recognize and respond to human emotions through body language [174], expressive movements [175], and verbal cues can enhance user satisfaction and foster cooperation [125].

Integrating these communication modalities enables quadruped robots to interact more effectively with humans, catering to different user preferences and situational requirements [176].

5.4.2. Safety and Ethical Considerations

Ensuring safety in human–robot interactions requires addressing both technical and ethical concerns [177]. Quadruped robots must operate reliably and ethically in order to earn user trust and comply with regulatory standards. One critical aspect of safety is collision avoidance, which involves the use of sensors and algorithms to prevent accidental contact with humans [124]. Real-time detection of people in the robot’s vicinity allows for dynamic path planning and speed adjustments, reducing the likelihood of accidents. Proximity sensors, vision systems, and machine learning algorithms are commonly employed to enhance the robot’s ability to navigate safely [178,179,180].

Adhering to regulatory compliance is another key factor in ensuring safety. Robots must meet established safety standards, such as ISO 13482 for personal care robots, which includes requirements for failsafes and emergency stop mechanisms [180,181]. Compliance with these guidelines helps to ensure that robotic systems operate within acceptable safety limits, thereby minimizing the risk of harm to humans.

Privacy concerns are also crucial, particularly when robots are equipped with cameras and microphones that collect sensitive data [182]. It is essential to responsibly handle this information in accordance with privacy laws and ethical guidelines. Implementing data encryption, secure storage, and user consent protocols can address potential privacy issues and protect users [183].

By prioritizing safety and ethical considerations, quadruped robots can operate more effectively in human environments while reducing risks and promoting positive, trustworthy interactions [184].

5.5. Enhancing Perception and Decision-Making in Robotics

Artificial intelligence significantly enhances quadruped robots’ ability to interpret sensory data and make informed decisions. Machine learning techniques enable robots to learn from experience, adapt to new environments, and perform complex tasks autonomously [126]. This section discusses how deep learning and adaptive behaviors contribute to the perception and decision-making capabilities of quadruped robots.

5.5.1. Deep Learning Techniques

Deep learning has transformed robotics by providing powerful methods for processing high-dimensional sensory data and learning complex mappings from inputs to outputs [185]. One of the key techniques is RL, which allows robots to learn optimal actions through trial and error while receiving feedback in the form of rewards or penalties from the environment [127]. In the context of quadruped robots, RL has been used to improve locomotion strategies and adaptation to varying terrains [186,187]. For instance, Hwangbo et al. demonstrated that quadruped robots could learn fast and robust locomotion policies in simulation, which could then be successfully transferred to real-world hardware [187].

Supervised learning is another critical deep learning technique. It is particularly useful for tasks where labeled data are available, such as object recognition and terrain classification [128]. By training models on large datasets, robots can achieve higher accuracy in perception tasks. CNNs have been widely adopted for visual perception in robotics, enabling robots to detect and classify objects in their environment, which is essential for navigation and manipulation tasks [110].

In addition to supervised learning, unsupervised learning plays a vital role in discovering patterns within sensor data without the need for labeled examples [129]. This approach is useful for tasks such as anomaly detection and clustering environmental features [188]. Techniques such as autoencoders and Generative Adversarial Networks (GANs) are employed to learn representations of sensory inputs [130], helping robots to extract useful features from raw data. Unsupervised learning reduces the need for manual labeling and enhances a robot’s ability to adapt to new and unfamiliar environments.

Collectively, deep learning techniques have significantly advanced quadruped robots’ capacity to process high-dimensional sensory data and make intelligent, context-aware decisions in real-time. These capabilities are essential for enabling robots to operate autonomously in complex and dynamic environments and to interact effectively with the physical world. In particular, neural network-based controllers—especially those trained through RL, Imitation Learning (IL), or hybrid paradigms combining the two—have demonstrated impressive performance in achieving agile, adaptive, and high-dimensional locomotion behaviors. However, a critical limitation of these approaches lies in their lack of formal guarantees regarding safety, robustness, and stability, especially when exposed to out-of-distribution scenarios or unexpected disturbances. Unlike classical model-based controllers, learned policies are often difficult to interpret or verify, making it challenging to ensure reliable deployment in safety-critical applications.

Recent efforts have begun to address this challenge through the integration of control theory and learning. One direction is the use of safe reinforcement learning, which incorporates constraints (e.g., via constrained Markov decision processes) to prevent unsafe actions during training and execution [189]. Another is learning with stability certificates, where the neural controller is trained alongside a Lyapunov function or by using control Lyapunov/barrier functions (CLF/CBF) to ensure provable safety and boundedness [190,191]. Furthermore, some works have employed formal verification techniques such as reachability analysis or SMT solvers to certify neural policies post-training, albeit with scalability challenges [192].

In the context of quadruped robots, ensuring that neural policies can handle contact-rich interactions and recover from perturbations without compromising safety is crucial. As such, the development of certifiable neural controllers is a pressing research direction, particularly for applications where failure may incur physical risk or damage, such as industrial inspection, human–robot collaboration, and search and rescue. Future research should aim to bridge the gap between high-performance learning-based locomotion and verifiable safety guarantees, enabling broader and safer adoption of intelligent quadruped robots in real-world scenarios.

5.5.2. Adaptive Behaviors

Machine learning enables quadruped robots to adapt to new situations and improve their performance over time. One key approach is online learning, which involves updating models in real time based on new data inputs [193]. This allows robots to continuously improve and adapt to changing environments. For quadruped robots, online learning is particularly useful for adjusting locomotion parameters on the fly, helping the robot to cope with unexpected terrain changes or sensor noise.

Another important technique is transfer learning, which involves applying knowledge gained in one task or environment to enhance performance in a different one [194]. This reduces the need for extensive retraining when deploying robots in different settings. For example, policies learned in simulation can be transferred to real robots, saving both time and resources [195]. Transfer learning enables rapid deployment and facilitates adaptation to new tasks or environments without the need for extensive recalibration.

Predictive modeling is also crucial for adaptive behaviors, as it allows the robot to anticipate future states of its environment or its own actions [196]. For example, predictive models can help a robot to foresee a slip when walking on a surface, enabling it to preemptively adjust its gait [197]. By anticipating potential issues, predictive modeling helps robots to plan and execute complex tasks more safely and efficiently, improving both performance and reliability.

Integrating these adaptive behaviors allows quadruped robots to operate more effectively in dynamic and unstructured environments, significantly enhancing their autonomy and robustness [126].

5.5.3. Embodiment and Bio-Inspired Architectures

An emerging perspective in quadruped control design is the notion of embodiment, which is the idea that intelligence emerges not only from computation but from the interplay between the control system, the robot’s body, and its environment [198]. In this context, embodied motor neuron architectures mimic the distributed and decentralized control strategies found in biological organisms, where local sensing and motor pathways collaborate to generate adaptive locomotion patterns.

Such architectures typically couple sensory feedback (e.g., proprioceptive and tactile signals) directly to motor outputs via modular neural units, often inspired by Central Pattern Generators (CPGs) or spinal cord circuits [199]. These systems can achieve robust rhythmic locomotion with limited computation and are capable of dynamically adapting to changing terrain, perturbations, or even hardware damage without explicit reprogramming.

For quadruped robots, integrating motor neuron-based controllers with learned policy modules or reflexive feedback loops has shown promise in enhancing both adaptability and energy efficiency. The tight coupling between morphology and control also facilitates morphological computation, where part of the control complexity is offloaded to the mechanical structure itself. These embodied strategies complement conventional learning-based control by providing inherently robust and reactive behaviors, especially in unstructured or unpredictable environments.

6. Applications Across Industries

Quadruped robots have found applications across a wide range of industries due to their advanced capabilities in mobility, adaptability, and interaction with complex environments. Their ability to traverse uneven terrains, avoid obstacles, and perform tasks autonomously makes them suitable for diverse sectors such as military, search and rescue, agriculture, and more. This section corresponds to RQ4: What are the key applications of quadruped robots across industries, and what challenges hinder their broader adoption?

6.1. Military and Defense

In the military and defense sectors, quadruped robots have become invaluable tools for navigating challenging environments and conducting operations that would otherwise pose significant risks to human soldiers. The development of quadruped robots such as Boston Dynamics’ BigDog was originally focused on load-carrying capabilities, allowing these robots to assist troops by transporting equipment across rugged terrains where wheeled vehicles cannot operate [12]. By reducing the physical burdens on soldiers, these robots enhance the efficiency of military logistics, especially in remote or hostile environments.

In addition to logistics, quadruped robots are increasingly being deployed in surveillance and reconnaissance missions. Equipped with advanced sensors and cameras, they can gather intelligence, monitor enemy movements, and relay real-time data without exposing human soldiers to danger [13]. Their ability to operate silently and move through difficult terrains makes them particularly effective for stealth operations and monitoring.

Another important application in the military is using quadruped robots in hazardous environments contaminated by Chemical, Biological, Radiological, or Nuclear (CBRN) agents. In these situations, quadruped robots can be deployed to perform tasks such as hazard detection, decontamination, and even material disposal, ensuring that human soldiers remain out of harm’s way [200].

6.2. Search and Rescue Operations

Quadruped robots have shown exceptional potential in the realm of Search and Rescue (SAR), particularly in disaster response scenarios. Their ability to traverse debris-strewn environments and access confined spaces makes them ideal for exploring disaster sites where traditional human responders may be hindered. For instance, quadruped robots can navigate through the rubble to locate survivors after earthquakes or building collapses, providing critical visual and auditory feedback to rescue teams [14].

In addition to locating survivors, these robots can deliver essential supplies such as medical kits, food, or communication devices to individuals trapped in areas inaccessible to humans. This capability is especially valuable in situations where human responders are unable to reach victims quickly due to structural instability or other dangers [15,16].

Moreover, quadruped robots can play a key role in assessing the structural integrity of buildings and infrastructure following disasters. Thanks to their specialized sensors, they can evaluate the safety of environments before human rescue teams enter, minimizing the risk of secondary collapses or other hazards [17,18]. Their autonomous operation and data collection capabilities make them essential tools for modern SAR operations.

6.3. Industrial Inspection and Maintenance

Quadruped robots are increasingly being adopted in industrial settings for inspection and maintenance tasks, particularly in hazardous or hard-to-reach areas. In the oil and gas industry, robots such as ANYmal have been deployed to inspect offshore platforms, pipelines, and storage tanks. By automating these processes, companies can reduce the need for human workers to enter dangerous environments, thereby improving safety and operational efficiency [201]. In power plants, quadruped robots are used to navigate complex industrial landscapes, monitor equipment, detect leaks, and measure thermal anomalies, providing real-time data that helps in predictive maintenance and reducing downtime [202]. Similarly, in manufacturing facilities, these robots can carry out routine inspections, monitor machinery for faults, and ensure compliance with safety regulations all while seamlessly integrating into existing workflows [203,204].

6.4. Agriculture and Farming

In the agricultural sector, quadruped robots are enhancing efficiency and productivity through automation and precision agriculture techniques. For instance, robots equipped with advanced sensors and cameras can be used for crop monitoring, assessing crop health, detecting diseases, and monitoring growth conditions [205]. This level of detailed monitoring allows farmers to make informed decisions about irrigation, fertilization, and pest control, leading to improved yields and resource management. Moreover, quadruped robots can play a role in livestock management; robots can herd animals, monitor their health, and ensure their wellbeing, all without causing stress to the animals, which is a common issue in traditional herding methods [206]. Additionally, they can perform soil analysis by collecting samples and analyzing soil properties, helping farmers to optimize their irrigation and fertilization strategies based on precise data [207].

6.5. Entertainment and Service Industries

Quadruped robots have also started making inroads into the entertainment, hospitality, and service sectors, offering new forms of interactive experiences. For example, robots such as Spot have been featured in artistic performances, dance routines, and public demonstrations, showcasing agility, coordination, and aesthetic appeal [208]. These performances not only entertain but also highlight the advanced capabilities of robotic systems in creative environments. In service industries such as malls and exhibitions, quadruped robots are being used to engage with customers, provide information, and even guide tours. Their ability to interact naturally with people makes them ideal for roles that require a balance of technical functionality and user engagement [209,210]. Furthermore, in education and research, quadruped robots serve as platforms for teaching robotics and programming, allowing students to engage with cutting-edge technology and encouraging interest in STEM fields [211].

6.6. Healthcare

While quadruped robots are still an emerging technology in healthcare, they increasingly show promise for enhancing efficiency and safety in medical environments. One application is patient assistance, where robots can transport supplies or medical equipment across hospital departments, reducing physical strain on human staff and optimizing workflows [212,213]. Furthermore, these robots can support rehabilitation by aiding patients during physical therapy sessions, providing stability and guidance as individuals work to regain mobility [214].

Sanitation is another important application, especially during health crises such as the COVID-19 pandemic. Quadruped robots can disinfect hospitals and other facilities, reducing the risk of pathogen transmission and creating a safer environment for patients and staff [215]. One promising use case is assisting visually impaired individuals in healthcare or general settings. Recent advancements have highlighted the development of quadruped robot systems such as RDog, which provides guidance to Blind and Visually Impaired (BVI) individuals. Equipped with advanced mapping and obstacle avoidance technologies, these robots can navigate diverse indoor and outdoor terrains. They enhance user safety and confidence through preemptive voice feedback and kinesthetic guidance, addressing limitations of traditional aids such as white canes and guide dogs [216]. The RDog system demonstrates improved navigation efficiency, reduced cognitive workload, and smoother guidance compared to traditional methods. These capabilities make quadruped robots an invaluable tool for improving accessibility and mobility in healthcare environments.

6.7. Environmental Monitoring and Conservation

In environmental conservation and monitoring, quadruped robots can contribute significantly to data collection and ecosystem management. For instance, they can be used for wildlife observation, allowing researchers to monitor animal populations and behaviors without disturbing their natural habitats. This non-intrusive approach provides valuable insights into species conservation efforts [217]. Additionally, quadruped robots are capable of tracking environmental parameters such as temperature, humidity, and pollution levels in remote or difficult-to-access areas. This allows for continuous monitoring of ecosystems and early detection of environmental changes [218,219]. In reforestation efforts, quadruped robots could assist in planting trees or dispersing seeds in areas that are challenging for humans to reach, accelerating the process of restoring damaged ecosystems [220].

6.8. Construction and Infrastructure

In the construction industry, quadruped robots are becoming valuable tools for site monitoring and safety assessments. These robots can navigate complex construction sites to monitor progress, check for compliance with building regulations, and identify potential hazards. In this way, they help project managers to ensure that work is being carried out safely and efficiently [221,222]. Furthermore, quadruped robots equipped with advanced technologies such as LiDAR can create detailed 3D maps of construction sites, which are useful for planning, analysis, and design purposes. This capability improves the precision and effectiveness of construction projects [223]. Additionally, quadruped robots can transport tools and materials to workers in areas where traditional vehicles cannot operate [224], enhancing overall workflow and reducing delays caused by logistical challenges.

6.9. Mining

Mining operations are often located in hazardous and complex environments, and as such stand to benefit from the deployment of quadruped robots. These robots can be used to inspect tunnels, check for structural integrity, and detect gas leaks or other dangers that might pose risks to human workers [225,226]. The ability of quadruped robots to navigate narrow and unstable tunnels makes them particularly suited for these tasks. Additionally, they can collect mineral samples from areas deemed unsafe for human access, minimizing the need for workers to enter dangerous zones [227,228]. Autonomous mapping is another important function, as quadruped robots can create detailed maps of underground mines, assisting in both safety planning and operational efficiency by providing real-time data on the conditions of the mines [229].

6.10. Space Exploration

Quadruped robots hold great potential for space exploration, where the ability to traverse uneven and unpredictable terrains is crucial. These robots could be deployed for planetary exploration missions, allowing them to navigate the surfaces of planets or moons, conduct scientific experiments, and collect soil or rock samples in locations that would be challenging for traditional rovers [230]. Their mobility and adaptability make them ideal for exploring diverse landscapes in space. Furthermore, quadruped robots could assist in constructing habitats or infrastructure for future human space missions, assembling components in environments where human involvement might be limited by gravity or other environmental factors [231,232]. Finally, they could serve as robotic assistants, helping astronauts by carrying equipment, performing maintenance tasks, or even conducting autonomous repairs on spacecraft or habitats [233].

6.11. Robots for Cultural Heritage Conservation and Archaeology

The preservation and exploration of cultural heritage sites and archaeological remains often require non-invasive and precise methods to mitigate the risk of damage. Robotics offers a transformative solution, enabling tasks such as exploration, documentation, restoration, and surveillance of culturally significant sites.

Robots equipped with advanced mobility systems [234], including legged or hybrid platforms such as HeritageBot III, can navigate the uneven and delicate terrain often found in ancient ruins and excavation sites. These systems merge ground locomotion with aerial capabilities to explore areas that are otherwise unreachable, thereby minimizing human intrusion. Advanced robotics platforms can carry sensors capable of performing non-invasive analysis such as spectroscopy or ground-penetrating radar to uncover hidden structures and assess the integrity of materials without physical interference.

Drones and legged robots [235] are commonly outfitted with LiDAR, high-resolution cameras, and 3D scanners, allowing them to generate accurate digital replicas of artifacts and structures. Such digital models are invaluable for archiving and virtual restoration while enabling global access to heritage through virtual reality.

Autonomous or semi-autonomous robots [236] can monitor the structural health of heritage sites by detecting changes or damage caused by environmental factors such as erosion, humidity, or temperature fluctuations. Periodic robotic surveillance ensures early detection of issues, reducing restoration costs. Robots equipped with manipulators can assist in precision tasks such as applying conservation materials or assembling fragmented artifacts. Their accuracy reduces the risk of human error and enables delicate operations in areas difficult for conservators to access.

7. Challenges and Limitations

Despite the significant advancements and potential applications of quadruped robots, several challenges (Figure 9) and limitations hinder their widespread adoption and deployment. This section addresses RQ5: What are the challenges, future directions, and potential research areas for advancing quadruped robotics?

7.1. Technical Challenges

7.1.1. Terrain Adaptability

Quadruped robots have demonstrated significant potential in navigating complex terrains, particularly compared to wheeled or tracked robots; however, several challenges remain, especially when operating in highly variable and extreme environments [42]. Quadruped robots must adapt to a wide range of surface conditions. For instance, loose gravel, mud, and ice present significant risks of slippage, which can lead to instability or falls. A robot’s ability to adjust its gait in real-time using proprioceptive and exteroceptive sensors is crucial for maintaining stability on such unpredictable surfaces [41]. Despite advancements in control algorithms, traversing these surfaces remains a challenge in terms of maintaining traction and balance [237]. In addition to surface variability, negotiating obstacles such as large rocks, gaps, and steep inclines poses difficulties. The limited length of a quadruped’s legs and the degree of joint articulation constrain its ability to overcome high obstacles or wide gaps. While advancements in leg morphology and joint actuation have enabled better terrain negotiation, the physical limitations of the robot’s structure still present challenges in extreme cases [238]. Moreover, environmental conditions such as heavy rain, snow, and dust storms can impact the robot’s sensor suite and mechanical components. Moisture and dust can interfere with sensors, reducing the accuracy of terrain mapping and foot placement planning [239]. Likewise, extreme cold and heat can affect battery life, motor efficiency, and joint flexibility, necessitating more robust hardware designs for extreme environments [240,241]. As sensor and mechanical technologies continue to evolve, overcoming these environmental challenges will be key to enhancing the terrain adaptability of quadruped robots.

7.1.2. Energy Consumption and Autonomy

Quadruped robots are often limited by their energy consumption, which directly impacts their operational autonomy. As robotic systems become more capable of complex locomotion and task execution, the demand for efficient power usage grows correspondingly [136]. One of the main constraints involves battery limitations. Together with onboard computing and sensor processing, the high-performance actuators required for dynamic and adaptive movements consume a significant amount of power. This results in relatively short operational periods before requiring recharging or battery replacement [41]. Current battery technologies are not yet optimized for long-term autonomy in mobile robots, limiting their deployment in extended missions, especially in remote or hostile environments. In addition, a tradeoff exists between weight and energy storage.Increasing battery capacity while extending operational time directly increases the robot’s weight. This additional weight can negatively affect mobility, reducing agility and increasing energy consumption during locomotion [242]. The challenge lies in finding a balance between sufficient energy storage and maintaining an efficient lightweight design that supports sustained operation. To address these issues, developing energy-efficient locomotion strategies is critical. Advanced control algorithms and hardware optimizations, such as using elastic components in actuators or designing energy recycling systems, have shown promise in reducing the overall power requirements of quadruped robots during motion [243]. Innovations such as these are key to improving mission duration and enabling robots to autonomously handle longer and more complex tasks.

7.1.3. Robustness and Reliability

Ensuring consistent performance across various operational conditions remains a significant challenge in quadruped robotics. Mechanical and environmental factors contribute to the difficulty of achieving reliable long-term operation [42]. One major concern is mechanical wear and tear. The repetitive stress on leg joints and actuators during locomotion can accelerate degradation, especially when navigating difficult or unpredictable terrain. Over time, these components are prone to failures that impact the robot’s performance and operational longevity [244]. Wear-resistant materials and designs that minimize friction are areas of active research aimed at extending the durability of these systems. Additionally, frequent maintenance requirements further complicate the deployment of quadrupeds in practical applications. Routine checks and part replacements are necessary to ensure continued reliability, which can be both costly and time-consuming. Maintenance in remote or hazardous environments can be logistically challenging, highlighting the need for more durable designs and self-diagnosis systems [245,246]. Finally, fault tolerance remains a crucial area of improvement. Currently, quadruped robots have limited ability to handle critical component failures, often leading to partial or total system shutdowns. Developing systems with redundancy and fault recovery mechanisms is essential for enhancing operational reliability, particularly in mission-critical applications [237]. Advances in control systems that allow robots to adaptively compensate for hardware malfunctions are pivotal in achieving robust autonomy.

7.1.4. Complexity of Control Systems

Advanced control algorithms are essential for enabling quadruped robots to operate autonomously in dynamic and unpredictable environments. However, these systems often require substantial computational resources, leading to performance bottlenecks in real-time applications [247,248]. The need for real-time processing is critical in quadruped robots, as perception, control, and decision-making must happen in a matter of milliseconds to maintain balance and react to environmental changes. High computational demands can result in latency, which in turn may affect the robot’s ability to respond to external stimuli promptly [249]. Reducing these latencies through hardware acceleration or optimized software architectures is a key area of ongoing research. Moreover, the algorithmic complexity involved in developing robust control systems capable of handling varied terrain and dynamic environments is a significant challenge. Balancing between sophisticated algorithms and computational feasibility is difficult, as these algorithms must consider factors such as stability, energy efficiency, and fault tolerance [136,250]. Simplified models may work well in controlled environments but fail in more complex scenarios, necessitating further improvements in adaptive control strategies. In recent years, machine learning has emerged as a promising tool for improving control and perception; however, it is not without limitations. Training data often fail to cover the full range of environmental variations that a quadruped might encounter, leading to unpredictable behavior when faced with unfamiliar conditions [251,252]. This limitation underscores the need for more generalized learning algorithms and simulation environments that can model a wider variety of scenarios.

7.2. Ethical and Social Considerations

As quadruped robots move from research laboratories to real-world applications, ethical and social issues have become increasingly important. These concerns range from safety risks to privacy issues and the broader societal impact of robot deployment.

7.2.1. Safety Concerns

The integration of quadruped robots into environments shared with humans introduces significant safety concerns. One major risk is collision, as robots can inadvertently harm individuals if they fail to navigate properly or to detect nearby humans. This risk is exacerbated by the increasing autonomy of robots, which reduces the direct control that humans have over their actions [253]. Another concern is unpredictable behavior resulting from software bugs, sensor failures, or inadequate response to environmental stimuli. Robots may malfunction or act erratically under certain conditions, posing a danger to nearby individuals. Robust safety protocols and fault-tolerant systems are critical to mitigating these risks [179,254]. In emergency handling situations, robots might not react as effectively as a human would. For instance, quadruped robots may require human intervention in order to make appropriate decisions in the event of a fire or other hazardous situations. Developing more sophisticated algorithms for handling emergency scenarios is an important step toward ensuring safer interaction between robots and humans [255].

7.2.2. Privacy and Surveillance

The deployment of quadruped robots equipped with cameras and sensors raises significant privacy concerns. These robots often collect large amounts of data about their environment, including video footage and environmental scans. If not properly regulated, this can result in unauthorized data collection where individuals or private property are recorded without consent [256,257]. The potential use of robots in surveillance activities also raises ethical questions. While robots could be beneficial for security purposes, they could also be misused to monitor individuals without their knowledge or permission. Thus, the issue of consent and data ownership becomes critical, particularly when these robots are deployed in public or semi-private spaces [258]. Furthermore, concerns over data security are growing, as the sensitive information collected by quadruped robots could be vulnerable to cyberattacks. Ensuring that data collected by robots is securely stored and transmitted is essential to preventing breaches that could expose personal information or sensitive surveillance data [259].

7.2.3. Job Displacement

Automation, including the deployment of quadruped robots, is likely to have a profound impact on employment, particularly in industries such as logistics, agriculture, and security. The issue of labor replacement is a major concern, as robots could take over tasks traditionally performed by humans, leading to job displacement [260,261]. A reduction in low-skill labor demand may lead to unemployment in certain sectors, disproportionately affecting workers who lack the skills necessary for higher-level positions. However, this shift also creates a demand for higher-skilled labor, particularly in the areas of robot maintenance, programming, and oversight. This transition necessitates retraining programs and educational initiatives to equip workers with the skills needed in a more automated economy [262]. The overall economic impact of robotic automation could widen economic disparities. Technological unemployment could exacerbate income inequality, as those with advanced skills and access to education benefit from the growing robot economy, while others are left behind. Policymakers must consider ways to mitigate these effects through social safety nets and workforce retraining initiatives [263].

7.3. Regulatory and Legal Frameworks

The rapid development of quadruped robots has created a gap between technological advancements and the regulatory frameworks needed to govern their safe and responsible deployment. As quadrupeds are increasingly used in public and private spaces, the absence of comprehensive regulations poses significant challenges [264].

7.3.1. Lack of Standards

The field of quadruped robotics suffers from a lack of operational guidelines for deployment, especially in public spaces. Unlike other technologies such as autonomous vehicles, there are no standardized protocols outlining how and where these robots can operate safely [265]. This regulatory vacuum means that decisions on safety, security, and ethical operation are often left to individual developers or organizations, leading to inconsistent practices. Moreover, there are currently no certification processes in place to assess the safety and reliability of quadruped robots before they are deployed. While some industries may adopt internal testing standards, a unified framework for certification, similar to aviation or automotive safety regulations, is currently lacking. This gap increases the risks associated with deploying robots in uncontrolled or unpredictable environments [266]. Another pressing issue is the question of liability. When accidents or damages occur due to robot malfunction, it is often unclear whether the robot manufacturer, operator, or owner is legally responsible [267]. This ambiguity could hinder the widespread adoption of quadruped robots, as organizations and users may be hesitant to assume legal risks without clear guidelines.

7.3.2. International Variations

The regulatory landscape for quadruped robots also varies significantly across countries, which complicates their global deployment. Import and export restrictions on robotic technologies can limit the transfer of advanced hardware and software components, especially when they are subject to export control regulations for dual-use technologies [268,269]. These restrictions slow down international collaborations and the spread of innovations. Furthermore, countries have different requirements for operational permissions when testing or deploying robots in public areas. While some countries may have clear pathways for obtaining such permissions, others may lack the infrastructure to regulate these technologies, resulting in bureaucratic delays and operational challenges [265]. Additionally, ethical guidelines around robot use often reflect cultural and societal differences. For example, attitudes toward surveillance, data privacy, and human–robot interaction can vary widely between regions, influencing what is considered acceptable use of quadruped robots [270]. These ethical differences necessitate region-specific regulations that can further complicate global deployment.

7.4. Cost and Accessibility

While quadruped robots have the potential to revolutionize various industries, their high cost remains a barrier to widespread adoption. The expenses associated with research, development, and production contribute to their limited accessibility, particularly for smaller institutions or developing nations [271,272].

7.4.1. High Development and Production Costs

The process of designing, prototyping, and refining quadruped robots is resource-intensive. Research and development costs account for a substantial portion of the overall expenditure, as innovative technologies in locomotion, perception, and control must be rigorously tested and optimized [41]. This process can take years, requiring significant investment in both human expertise and advanced equipment. In addition, material and component costs significantly contribute to the high price of quadruped robots. High-quality materials that can withstand harsh environmental conditions are often expensive, as are precision-engineered components such as actuators and sensors. This limits the ability to scale up production and drives up the price of individual units [273,274]. The absence of economies of scale further exacerbates the cost issue. Quadruped robots are still produced in relatively low volumes; without large-scale manufacturing, costs remain high. Mass production could lead to cost reductions; however, achieving scale requires broader market adoption, which is difficult given the current price point [275].

7.4.2. Limited Accessibility for Research and Education

The high costs of quadruped robots also restrict their accessibility for educational use. Many schools and universities cannot afford to purchase these robots for teaching and research purposes, particularly those with limited budgets. This limits opportunities for students to gain hands-on experience with advanced robotic systems, creating a skills gap in the next generation of engineers and roboticists [276,277]. Research limitations due to budget constraints also hinder innovation. Smaller research institutions or labs in developing countries may not have the resources to purchase or maintain quadruped robots, limiting their ability to experiment with new ideas or contribute to the broader field [278]. As a result, the advancement of quadruped robotics could become increasingly concentrated in well-funded organizations, slowing the pace of progress. While some open-source options exist for quadruped robotics, they are often limited in terms of functionality or scalability. The lack of affordable high-quality platforms restricts collaborative development efforts, as researchers and educators are unable to access or modify the latest technologies [279,280]. Expanding the availability of open-source quadruped platforms could democratize access to these systems and accelerate innovation.

7.5. Environmental Impact

The environmental impact of quadruped robots is an increasingly important consideration as their deployment expands. From energy consumption to their effects on natural ecosystems, robots must be designed with sustainability and minimal environmental disruption in mind [281].

7.5.1. Energy Consumption

One of the primary environmental concerns associated with quadruped robots is their energy consumption. These robots often require high amounts of power to operate efficiently, especially when performing tasks such as locomotion over rough terrain or carrying heavy loads [237]. When this energy is sourced from non-renewable fossil fuels, it contributes directly to the robot’s carbon footprint. Given the global push toward reducing carbon emissions, integrating renewable energy solutions such as solar-powered charging stations could help to mitigate this issue [282]. In addition to energy use, battery disposal poses another environmental challenge. Many quadruped robots rely on lithium-ion or similar battery technologies, which can release harmful chemicals into the environment if not disposed of properly. Ensuring that battery recycling processes are in place is essential to reducing the ecological harm associated with large-scale robotic deployments [283]. The resource consumption during the manufacturing process of quadruped robots also requires consideration. The extraction of raw materials such as rare earth metals for electronics and actuators together with energy-intensive manufacturing processes contribute to these robots’ overall environmental footprint. Developing more sustainable materials and energy-efficient manufacturing techniques will be critical for reducing the long-term environmental impact of quadruped robots [284,285,286].

7.5.2. Wildlife Disturbance

Another environmental concern is the potential wildlife disturbance caused by the deployment of quadruped robots in natural environments. As these robots are increasingly used for environmental monitoring or search and rescue tasks, their presence may disrupt local ecosystems [287]. Robots could cause behavioral changes in wildlife. For example, robots’ unfamiliar sounds and movements may induce stress or cause animals to flee their natural habitats. This could lead to changes in feeding, mating, or migration patterns, with potentially long-term ecological consequences [288,289]. Furthermore, the physical presence of robots in sensitive ecosystems could lead to habitat disruption. Traversing through forests, wetlands, or other fragile environments may result in soil compaction, plant damage, or the displacement of small organisms. Minimizing the ecological footprint of robots through careful planning and deployment strategies is vital for mitigating such impacts. Noise pollution is another factor to consider [290]. The mechanical noises generated by actuators, motors, or cooling systems could disturb local fauna, particularly in quiet environments. Reducing operational noise through improved designs and quieter materials will be essential to ensuring that quadruped robots can operate in harmony with their surroundings.

7.6. Security Vulnerabilities

As quadruped robots become more integrated into various sectors from industrial applications to public service, their susceptibility to security threats is increasing. These robots are often equipped with advanced sensors and communication systems, making them attractive targets for cyberattacks [291].

7.6.1. Cybersecurity Risks

As with other autonomous systems, quadruped robots are vulnerable to cybersecurity risks. One of the most concerning threats is unauthorized access, where hackers could take control of the robot and use it for malicious purposes. Such breaches could have severe consequences, ranging from sabotage of industrial operations to threats to public safety in environments where robots interact with humans [291]. Additionally, robots often collect large amounts of sensitive information, such as environmental data, surveillance footage, or operational logs. In the event of a data theft, this information could be compromised, leading to privacy violations or the exposure of sensitive industrial processes [292]. As the use of quadruped robots expands into areas such as border patrol, law enforcement, and private security, safeguarding the data they collect will become paramount. The risk of system disruption through cyberattacks is also significant. Malicious actors could disable robots and render them inoperative, or interfere with their decision-making processes to cause them to malfunction. Such attacks could lead to financial losses, operational delays, or even dangerous situations in mission-critical environments [293]. Building robust cybersecurity defenses, including encrypted communication and real-time threat detection, is crucial for ensuring the safe operation of quadruped robots.

7.6.2. Dependence on Communication Networks

Many quadruped robots rely heavily on communication networks for their operation, especially when remote control or cloud-based data processing is involved. This dependence introduces additional vulnerabilities. For instance, signal interference can disrupt the communication between the robot and its operator or central processing system, potentially affecting performance. In hostile environments or areas with poor signal coverage, this could lead to operational failures [294]. Jamming attacks are another serious threat; malicious actors could intentionally interfere with communication signals, rendering the robot inoperative. Such attacks could be particularly harmful in critical applications where continuous communication is vital, such as disaster response or military operations [295,296]. Finally, network latency challenges real-time control and decision-making. Even small delays in data transmission can affect robots’ ability to respond to environmental changes or execute complex maneuvers. Ensuring reliable low-latency communication networks, possibly through 5G or dedicated satellite connections, will be important for improving the performance and security of quadruped robots [297].

8. Future Perspectives and Research Directions

The development of quadruped robotics is advancing at a rapid pace, with ongoing research addressing current limitations and pushing the boundaries of what these robots can achieve. This section further addresses RQ5: What are the challenges, future directions, and potential research areas for advancing quadruped robotics? Figure 10 shows a timeline highlighting some enhancements of quadruped robots that are expected in the near and far future.

8.1. Advancements in Artificial Intelligence and Machine Learning

It is anticipated that AI and ML [8,139] will be central in driving increased autonomy and adaptability of quadruped robots. AI techniques, especially RL [86], are being leveraged to optimize robot locomotion through self-improvement in simulated environments. RL systems enable robots to not only learn from their environments but also generalize to new unseen terrains with minimal human intervention. Furthermore, transfer learning [194] methods help to bridge the gap between simulation and real-world applications, reducing dependence on extensive physical trials. As robots face dynamic and uncertain environments, adaptive behavior algorithms will allow them to respond in real-time when adjusting to obstacles and changes in terrain conditions, thereby enhancing their robustness and operational flexibility.

8.2. Integration of Soft Robotics

Integrating principles from soft robotics [285] holds the potential to significantly improve the performance, adaptability, and safety of quadruped robots. The use of compliant materials in the construction of limbs and joints can facilitate safer interactions with humans and environments, reducing the risk of damage or injury. Additionally, bio-inspired designs are being investigated to emulate the efficiency and fluidity of natural movements found in animals. Variable stiffness mechanisms, which allow actuators to transition between rigid and flexible states, offer a hybrid approach that enables both precision control and shock absorption as required. This adaptability in design fosters enhanced movement efficiency across a variety of terrains and operational contexts.

8.3. Enhanced Human–Robot Interaction

A critical area of research is improving human–robot interaction [122,124,184], particularly for collaborative tasks. As robots become more integrated into human-centric environments, intuitive interfaces such as gesture control and natural language processing are being developed to simplify user interactions with robots. These interfaces aim to make the operation of quadruped robots more accessible to non-experts, fostering broader adoption in commercial and personal applications. Social robotics research is also contributing by embedding social cues and behaviors into quadruped robots, enabling smoother interactions and increasing user trust [298,299]. Safety protocols remain a key consideration, with an emphasis on the development of robust systems to prevent accidents and ensure that robots can operate safely in shared spaces with humans [124,300].

8.4. Swarm Robotics and Collaborative Behaviors

Coordinating multiple quadruped robots [133] opens up new possibilities in tasks such as search and rescue, environmental monitoring, and agriculture. Research into swarm robotics is focused on creating effective algorithms for multi-robot coordination [301,302], allowing these systems to work together efficiently [303,304]. By sharing sensor data and creating distributed environmental maps [305,306], swarms of robots can cover larger areas more quickly and provide comprehensive situational awareness. Collective behaviors [307] inspired by biological systems such as insect colonies offer new avenues for robots to perform complex tasks through simple interaction rules, leading to emergent capabilities that surpass the sum of individual robots’ abilities.

8.5. Advancements in Materials and Fabrication

The continued development of advanced materials and fabrication techniques [55,283,285] is poised to revolutionize quadruped robotics. New composite materials that offer high strength-to-weight ratios are allowing the creation of lighter and more durable robots capable of withstanding harsher environments. Additive manufacturing (3D printing) is playing a critical role in this evolution by enabling the creation of custom components with complex geometries, which are otherwise difficult or impossible to produce using traditional methods. Additionally, smart materials such as shape-memory alloys offer promising applications, allowing robots to adapt their structure in response to external stimuli for improved versatility and durability.

8.6. Energy Efficiency and Power Management

Energy efficiency remains a pivotal challenge in quadruped robotics, especially for extended field operations. Current research is exploring innovative methods to improve energy management, including energy harvesting technologies that capture ambient energy from sources such as solar power or kinetic movements. Advancements in battery technology [75] are also crucial to extending operational times, particularly in developing high-capacity and fast-charging solutions. Furthermore, optimizing locomotion through efficient gait patterns and control algorithms will be essential to achieving reduced energy consumption, particularly during prolonged operation in remote or energy-scarce environments.

8.7. Regulatory Development and Standardization

As quadruped robots become more prevalent, establishing regulatory frameworks and standards to ensure their safe and ethical deployment is increasingly important. Safety standards must be developed in order to guide the design and operation of these robots, particularly when used in public or industrial environments. Ethical considerations around issues of privacy and surveillance must also be addressed to ensure responsible use. Furthermore, interoperability standards will be critical to ensuring that robots from different manufacturers can communicate and work together, fostering a more collaborative and unified robotics ecosystem.

8.8. Commercialization and Market Trends

The commercial viability of quadruped robots is growing as production costs decrease and technological capabilities expand. Cost reductions driven by advances in materials, manufacturing processes, and AI are making these robots more accessible across industries. New business models such as Robots as a Service (RaaS) [308] are emerging as well, allowing businesses to deploy robots without the high upfront costs traditionally associated with robotic systems. As their capabilities and affordability increase, the market for quadruped robots is expected to expand into sectors such as logistics, healthcare, and consumer markets.

8.9. Interdisciplinary Research and Collaboration

The future of quadruped robotics will rely on interdisciplinary collaboration [309], drawing insights from fields such as biology, neuroscience, materials science, and computer science. This cross-pollination of ideas will drive innovation and enable the development of more advanced and capable robotic systems. Open-source platforms are also critical in fostering collaboration, allowing researchers and developers to share their work and build on each other’s progress. Additionally, educational initiatives that integrate robotics into curricula will be essential in training the next generation of engineers and researchers, ensuring continued growth and innovation in the field.

9. Conclusions

Quadruped robots have progressed significantly, evolving from early prototypes that established basic principles of legged locomotion to sophisticated platforms capable of complex tasks. Their unique mobility, adaptability, and versatility position them as valuable tools in environments that are too challenging or dangerous for humans. Advances in mechanical design, intelligent control strategies, and sensor integration have enabled reliable navigation and dynamic interaction with the world. Current applications span a wide range of domains, including military, search and rescue, inspection, agriculture, healthcare, entertainment, environmental monitoring, construction, mining, and space exploration. As the field matures, key challenges such as improving terrain adaptability, increasing energy efficiency, and addressing ethical and regulatory concerns must be overcome. Looking ahead, ongoing research into artificial intelligence, soft robotics, materials science, human–robot interaction, and swarm collaboration will continue to shape the development of quadruped robots. Through interdisciplinary efforts and responsible governance, these systems will evolve to ensure their beneficial integration and ethical deployment within society.