A Survey on Object-Oriented Assembly and Disassembly Operations in Nuclear Applications

Abstract

Nuclear environments demand exceptional precision, reliability, and safety, given the high stakes involved in handling radioactive materials and maintaining reactor systems. Object-oriented assembly and disassembly operations in nuclear applications represent a cutting-edge approach to managing complex, high-stakes operations with enhanced precision and safety. This paper discusses the challenges associated with nuclear robotic remote operations, summarizes current methods for handling object-oriented assembly and disassembly operations, and explores potential future research directions in this field. Object-oriented assembly and disassembly operations are vital in nuclear applications due to their ability to manage complexity, ensure precision, and enhance safety and reliability, all of which are paramount in the demanding and high-risk environment of nuclear technology.

1. Introduction

Fusion power could be one of the very few sustainable options to replace fossil fuels as the primary energy source [1]. Unlike traditional nuclear fission, which generates energy by splitting atoms, fusion power harnesses the energy released when atomic nuclei fuse together, replicating the process that powers the sun and stars [2]. This process has the potential to produce virtually limitless energy from isotopes of hydrogen, such as deuterium and tritium. Fusion offers numerous advantages over other energy generation methods, including low greenhouse gas emissions, lower volumes of long-lived radioactive waste compared with fission energy, and a virtually inexhaustible fuel supply.

Despite its immense potential, fusion power also faces significant technical and practical challenges. One of the primary challenges is achieving and sustaining the extreme conditions required for fusion reactions to occur, including high temperatures and pressures [3]. Containing and controlling the plasma, the superheated gas where fusion occurs, within a magnetic field is particularly challenging [4]. Additionally, developing materials that can withstand the intense heat and radiation produced by fusion reactions over prolonged periods is an essential outstanding challenge [5]. Another challenge is the complexity and cost of building and operating fusion reactors, which require sophisticated engineering and technology.

Plant maintenance and decommissioning are essential aspects of managing and sustaining fusion energy facilities. Maintenance involves regular upkeep, inspection, and repair of fusion reactors and associated infrastructure to ensure their safe and efficient operation [6]. This includes tasks such as monitoring equipment performance, replacing worn components, and addressing any issues that may arise to prevent disruptions in operation. Additionally, decommissioning plays a crucial role in the lifecycle of fusion facilities, involving the safe dismantling and disposal of equipment and materials at the end of their operational life [7]. This process requires careful planning and execution to manage radioactive waste and minimize environmental impact.

Fission and fusion are two fundamental nuclear processes that release vast amounts of energy by altering atomic nuclei. The connection between fission and fusion lies in their complementary roles within nuclear science and energy production. While fission has been mastered and widely used, fusion remains the ultimate goal in energy research. Plant maintenance represents a critical aspect of ensuring the sustained operation and efficiency of fusion facilities, which hold immense promise for clean and abundant energy generation. The Joint European Torus (JET) [8] and International Thermonuclear Experimental Reactor (ITER) [9] represent two significant milestones in the pursuit of practical fusion energy. Both facilities play pivotal roles in advancing fusion research, with JET (Figure 1) serving as a crucial experimental platform and ITER poised to demonstrate sustained fusion reactions on a commercial scale. Maintenance of these facilities is paramount to ensuring their continued operation and progress towards achieving fusion energy. The plant maintenance activities at ITER will encompass a wide range of tasks, including regular inspections, equipment upgrades, and repairs to ensure the facilities’ optimal performance [10,11]. However, this maintenance presents unique challenges, particularly in integrating advanced technologies like robot manipulators. Maintaining and repairing these systems require specialized knowledge and skills, especially when employing robot manipulators for tasks in high-radiation and extreme-temperature environments. Developing and implementing a remote maintenance system poses a significant challenge for the ITER project, requiring the system to ensure high Tokamak availability while adhering to the overarching objectives of the ITER program [12]. Additionally, fusion facilities operate under extreme conditions, including high temperatures, intense magnetic fields, and exposure to radiation, which pose significant challenges for maintenance personnel and equipment. These conditions necessitate the development of robust robot manipulators capable of withstanding such environments while performing delicate and precise tasks.

When a fusion reactor eventually reaches the end of its operational life, the decommissioning process begins. Decommissioning represents a critical phase in the lifecycle of nuclear facilities, involving the safe and efficient dismantling of infrastructure, decontamination of radioactive materials, and management of nuclear waste. Sellafield [13], the UK’s largest nuclear site, is undergoing a significant decommissioning process to address its legacy of nuclear operations and waste management. This decommissioning effort involves the safe dismantling and disposal of obsolete nuclear facilities, as well as the management of radioactive waste and contaminated materials [14]. As a specific example, the process for decommissioning and dismantling gloveboxes, which are used for handling hazardous materials [15,16], involves removing any residual radioactive materials, decontaminating the equipment and surrounding areas to reduce radiation levels, and finally dismantling the boxes themselves. In JET, the dismantling process of the vacuum vessel and tritium plant will generate a significant amount of metallic waste, a considerable portion of which will be classified as Intermediate Level Waste (ILW) due to its tritium and/or activation levels [17]. One of the foremost challenges of decommissioning lies in the inherent complexity of nuclear facilities, characterized by intricate designs, tight spaces, and high radiation levels. Robotic systems deployed for decommissioning must navigate these environments with precision and reliability, often in conditions where direct human intervention is not feasible due to safety concerns. Moreover, ensuring the integrity of decommissioned components and preventing the spread of contamination poses significant challenges. Robotic manipulators must be capable of delicately handling radioactive materials and equipment while minimizing the risk of further contamination. Developing innovative robotic technologies is imperative to address these challenges and ensure the successful and safe decommissioning of nuclear facilities.

Future fusion plants face several challenges and limitations that hinder their widespread adoption and large-scale commercialization. One of the primary issues is their high maintenance demands, driven by the complexity of their systems and the extreme conditions in which they operate. The intense heat, radiation, and strong magnetic fields generated by fusion reactions cause significant wear and tear on critical components, requiring frequent inspections and repairs to ensure safe and efficient operation. Robotic systems used in nuclear applications must meet stringent requirements that commercial systems typically do not, primarily due to the extreme and hazardous environments in which they operate. These robots must withstand high levels of ionizing radiation, which can damage or degrade conventional electronic components and materials. They are also required to function reliably in confined, cluttered, or underwater environments, often with limited or no direct human oversight. Additionally, nuclear robots must exhibit high levels of precision, durability, and fail-safe mechanisms to handle delicate tasks like decommissioning, inspection, and maintenance of radioactive facilities, where errors could have severe safety and environmental consequences. Traditional maintenance methods, which rely heavily on human intervention via telemanipulation, will be inadequate due to the dangers and limitations of accessing these environments. Autonomous planning and execution are essential for addressing these challenges, enabling the development of intelligent robotic systems capable of performing regular inspections, detecting potential issues early, and executing precise repairs [18]. By incorporating autonomous planning and execution, future fusion plants could minimize downtime, reduce operational costs, and enhance safety by limiting human exposure to hazardous conditions [19]. This would ultimately ensure long-term reliability and performance as future fusion plants strive to deliver scalable, cost-effective, and reliable energy solutions, reducing the need for human involvement and improving overall efficiency.

This survey examines nuclear robotic remote operations and their associated challenges, summarizes the current technologies in automation that could be used for future plant maintenance and decommissioning, highlights various prototypes developed for assembly and disassembly operations, and identifies promising future research directions.

2. Nuclear Robotic Remote Operations and Their Challenges

2.1. Remote Inspection

Nuclear remote inspection (Table 1) involves the use of advanced technology and robotics to conduct inspections of nuclear facilities and infrastructure without direct human intervention. This is employed in environments where radiation levels are high, or access is restricted for other reasons, ensuring the safety of inspection personnel while effectively assessing the condition of nuclear assets. Remote inspection techniques include the use of cameras, sensors, and robotic manipulators to gather data on equipment condition, performance, and structural integrity, as well as any potential hazards. AutoInspect [20], a platform designed for robust and scalable mission-level autonomy, was deployed in JET for weeks. It integrated reliable mapping, as well as localization with autonomous navigation, scheduling, and mission execution, delivering a fully autonomous remote inspection system. Remote inspection at ITER will involve the use of advanced robotic systems equipped with cameras, sensors, and manipulators to gather data on equipment performance, structural integrity, and potential hazards. The manipulator (Figure 2a) developed by the Interactive Robotics Unit of CEA-LIST was introduced in [21] to deal with the problem of close inspection intervention tasks in a Tokamak. An articulated inspection arm robot was proposed in [22] to solve the potential problem of ITER inspection of the first wall and divertor cassettes. Remote inspection will allow ITER operators to monitor the condition of critical components such as the vacuum vessel, plasma-facing materials, and other key systems, enabling them to detect anomalies, plan maintenance activities, and ensure the safe and efficient operation of the fusion reactor. Remote inspection can aid in detecting anomalies, identifying areas for maintenance or repair, and monitoring changes over time. The integration of remote inspection technology enhances efficiency, reduces downtime, and minimizes the risk of human exposure to radiation, contributing to the safe and reliable operation of nuclear facilities.

Figure 2. Long-reach manipulators for remote inspections. (a) The manipulator developed in CEA-LIST Lab [21]. (b) TARM (Telescopic Articulated Remote Mast) [23].

Figure 2. Long-reach manipulators for remote inspections. (a) The manipulator developed in CEA-LIST Lab [21]. (b) TARM (Telescopic Articulated Remote Mast) [23].

Mobile robots [24,25] play a crucial role in nuclear remote inspection since they can be utilized to navigate through complex and hazardous areas within nuclear facilities. For instance, Spot (Figure 3) [26] is able to traverse challenging terrain and access hard-to-reach areas within nuclear facilities because of its agile and adaptable design, which could help with nuclear decommissioning. A robot, Spider (Figure 4a), was developed in [27] to deal with the problem of accessing unstructured areas on the Sellafield site. The Aqua Vehicle Explorer of In Situ Sensing (AVEXIS) vehicle (Figure 4b) was introduced in [28] for the purpose of monitoring legacy storage ponds on the Sellafield site. By leveraging autonomous navigation capabilities, mobile robots can autonomously explore and inspect various components and structures, collecting data on their condition and integrity. In [27], the Mini Robots for Restricted Access Exploration (MIRRAX) project was investigated aiming at remote inspection of dry storage facilities with restricted access. Mobile robots can be equipped with cameras, as well as temperature and radiation sensors, for performing tasks such as sampling, measurements, and visual assessments. Overall, using mobile robots for nuclear remote inspection enables thorough assessments of critical infrastructure while mitigating risks to human operators.

Despite its advantages, nuclear remote inspection [29] poses several challenges. One significant challenge is developing robotic systems capable of operating effectively in the harsh and complex environments prevalent within nuclear facilities. These environments often feature high radiation levels, extreme temperatures, and confined spaces, which can affect the performance and reliability of robotic equipment. Radiation can significantly affect the performance of robots by disrupting both their hardware and software systems. High-energy particles may cause single-event upsets in electronic components, leading to bit flips in memory and processors that result in data corruption or control errors. Prolonged exposure can degrade sensors and semiconductors, reducing their reliability and lifespan. These hardware issues often lead to faulty data interpretation, navigation errors, and control instability within the robot’s decision-making systems. Ensuring the accuracy and reliability of inspection data collected remotely is another challenge, as it requires sophisticated sensors and imaging technologies capable of providing detailed and accurate information about the condition of nuclear assets. Furthermore, the integration of remote inspection systems with existing infrastructure and operations can be complex, requiring careful planning and coordination to ensure compatibility and effectiveness. Within the remote handling control room of JET, inspection specialists meticulously assess various factors during operations. These include verifying the aspect ratio of targets in each shot, addressing potential lens flare that might obstruct readings, and identifying natural reflections that could be misconstrued by the software as retro-reflective targets [30]. Recent advancements in design and technology present an opportunity for remote handling operators to assume a more active role in the process since this allows for a simplified operational interface, enabling inspection specialists to adopt a more supervisory position, overseeing the proceedings with enhanced efficiency [30,31]. As a result, remote inspection may face limitations in certain scenarios where direct human intervention is necessary, such as in tasks requiring complex decision-making or dexterity.

Table 1. Summary of nuclear remote inspection technologies and challenges.

Category	Details
Purpose	Inspect nuclear facilities remotely in hazardous, high-radiation, or restricted environments.
Technologies	Cameras, sensors, robotic manipulators, mobile robots, autonomous navigation, and scheduling systems.
Key Platforms/Examples	AutoInspect (JET) [20], CEA manipulator [21], and TARM (Figure 2b) [23].
Mobile Robots	Spot [26], Spider [27], MIRRAX [27], and AVEXIS [28].
Applications	Inspection of vacuum vessels, plasma-facing materials, storage ponds, or dry storage facilities.
Challenges	Harsh environments (radiation, temperature, confined spaces), data accuracy, integration complexity, and need for occasional human intervention.
Recent Advancements	Simplified control interfaces, enhanced supervisory roles, and improved autonomy and mission-level control.
Limitations	Tasks needing complex human judgment or dexterity remain difficult for fully remote systems.

Table 1. Summary of nuclear remote inspection technologies and challenges.

Category	Details
Purpose	Inspect nuclear facilities remotely in hazardous, high-radiation, or restricted environments.
Technologies	Cameras, sensors, robotic manipulators, mobile robots, autonomous navigation, and scheduling systems.
Key Platforms/Examples	AutoInspect (JET) [20], CEA manipulator [21], and TARM (Figure 2b) [23].
Mobile Robots	Spot [26], Spider [27], MIRRAX [27], and AVEXIS [28].
Applications	Inspection of vacuum vessels, plasma-facing materials, storage ponds, or dry storage facilities.
Challenges	Harsh environments (radiation, temperature, confined spaces), data accuracy, integration complexity, and need for occasional human intervention.
Recent Advancements	Simplified control interfaces, enhanced supervisory roles, and improved autonomy and mission-level control.
Limitations	Tasks needing complex human judgment or dexterity remain difficult for fully remote systems.

2.2. Remote Handling

Nuclear robotic remote handling (

Table 2. Summary of nuclear robotic remote handling systems and challenges.

Aspect	Details
Purpose	Remote handling of radioactive components, infrastructure maintenance, and inspections in hazardous nuclear environments.
Key Systems	MASCOT: Bilateral force feedback telemanipulator with seven DoFs per arm, used for over 30 years in JET [33,34]. MASCOT 6: Modernized version with updated electronics and control systems [35]. Telbot: Telemanipulator for dexterous tasks via teleoperation [36,37].
Capabilities	Advanced manipulators, sensors, cameras, force feedback, and autonomous or teleoperated systems for inspection, lifting, and material handling.
Major Projects	JET, ITER, and DEMO.
Challenges	Withstanding high radiation, temperature, and confined spaces. Integrating complex manipulator systems with plant infrastructure. Ensuring inspection data accuracy and operational safety. Developing intuitive human–robot interfaces and autonomous planning. Adapting to unknown or changing plant conditions.
Advancements	Modernization of manipulators (MASCOT 6). Integration of haptic feedback and advanced teleoperation interfaces. Progress in autonomy and mission-level control strategies.

) involves the use of robotic tools to perform tasks in hazardous nuclear environments, such as replacing radioactive components, maintaining infrastructure, and conducting inspections. Utilizing advanced robotic systems equipped with specialized sensors, manipulators, and cameras, nuclear remote handling enables remote operation and maintenance of nuclear facilities. A major output from the JET project is the practical demonstration of the feasibility of remote handling for a Tokamak [32]. MASCOT (Figure 5) [33], a bilateral force feedback manipulator, has been used for over 30 years to carry out maintenance activities within the JET vessel, minimizing human access to the vessel whilst improving maintenance tasks and tools using the telemanipulation system and its productivity over time. This long-term effort has enabled the manipulator to be equipped with various lifting devices, thereby enhancing its flexibility, operability, and overall dexterity [33]. The MASCOT telemanipulator [34] serves as the primary tool for remote handling operations in JET. Selected by the remote handling group in the late eighties, MASCOT was designated to establish the foundation of the remote maintenance system for the JET torus [34]. This two-arm Master–Slave device has seven degrees of freedom (DoFs) per arm, including a gripper, enhancing its versatility and functionality [34]. Later, from 2014 onwards, the RACE engineers have started working on MASCOT 6, which is an upgraded design of the original hardware and focuses on designing the telemanipulator with modern electronics and control systems. MASCOT has earned its reputation through thousands of hours of successful operation inside JET Tokamak [35]. Its reliability and effectiveness in handling complex remote maintenance tasks make it an ideal choice for fusion devices like JET, ITER, and DEMO in the future [35], although an increased level of automation and minimal human intervention with telemanipulators are desired. Remote handling robots are equipped with sensors, manipulators, and cameras to perform tasks remotely, keeping human operators safe from radiation exposure.

Telemanipulation (Figure 6) is a specialized form of remote handling that involves the use of robotic manipulators to perform tasks in hazardous or inaccessible environments [36]. Unlike purely mechanical remote handling tools, telemanipulation systems are more complex and include advanced manipulators that replicate human movements with precision. These systems allow operators to control robotic arms and tools from a safe distance, providing a high level of dexterity and control for tasks such as maintenance, inspection, and material handling [37]. Telemanipulation often incorporates feedback mechanisms, such as haptic or force feedback, in order to give operators a sense of touch and enhance their control over the manipulator [38]. This technology is essential in industries like nuclear energy, where direct human intervention may be too risky or impossible. By combining advanced robotics and intuitive control interfaces, telemanipulation systems offer a safer and more efficient way to perform complex tasks in challenging environments.

Nuclear remote handling presents a host of challenges that must be addressed to ensure the safe and efficient operation of nuclear facilities. One significant challenge is developing robotic systems capable of withstanding the harsh conditions prevalent within nuclear environments, including high radiation levels, extreme temperatures, complex geometries, etc. These robots must be highly reliable and durable to perform tasks such as handling radioactive materials, conducting inspections, and performing maintenance. Maintenance and repair of robotic systems deployed in highly radioactive and contaminated environments present several significant challenges. Direct human intervention is often impossible due to hazardous radiation levels, making remote, automated, or teleoperated repair methods essential yet technically complex. Components may degrade rapidly under radiation exposure, causing unexpected failures in electronics, sensors, and mechanical parts, while contamination limits the safe retrieval, handling, or disposal of damaged equipment. Access to confined, underwater, or structurally unstable areas further complicates repair operations. Additionally, designing modular, easily replaceable parts that can be swapped using remote tools is difficult, and ensuring long-term system reliability with limited maintenance opportunities requires careful material selection, redundant systems, and rigorous fault-tolerant designs. The major challenges of the overall ITER project include the need for timely, safe, and effective remote operations in such a nuclear environment and the novelty and complexity of the remote handling requirements, which make the ITER remote maintenance system a key component in ITER’s design and operation [39]. Additionally, designing intuitive human–robot interfaces is essential to enable operators to control and monitor robotic actions effectively from a remote location. Another challenge is ensuring the accuracy and reliability of inspection data collected remotely, as this information is crucial for assessing the condition of nuclear assets and detecting anomalies. Furthermore, enhancing the autonomy of robotic systems to adapt to dynamic environments and perform complex tasks independently remains a key challenge. There is a need for ongoing research and development to enhance the autonomy and dexterity of nuclear robots, allowing them to adapt to unforeseen situations and perform complex tasks with minimal human intervention. The challenge faced by the JET project was to establish a remote handling capability that seamlessly incorporated elements of Tokamak design and construction, remote manipulation devices, and remote operations facilities within an environment characterized by uncertain plant configurations and conditions in both the short- and long-term [40]. The execution of remote operations necessitated meticulous integration of manipulation devices with suitably prepared plant components and standardized operation techniques [41]. Overcoming these challenges is vital for the advancement of nuclear robotic remote handling technology, enhancing safety and efficiency in nuclear operations, and fostering autonomy and autonomous planning in robotic systems.

2.3. Autonomous Planning and Execution

Future fusion plants would face significant challenges in terms of maintenance due to the extreme and hazardous environments they operate in. The harsh environment of nuclear-related tasks significantly influences the design and functionality of robotic technologies by demanding enhanced durability, reliability, and adaptability. Exposure to high radiation levels, extreme temperatures, corrosive substances, and confined or underwater spaces requires robots to be built with radiation-hardened materials, robust shielding, and specialized components resistant to environmental degradation. Functionality must emphasize remote operation, precise manipulation, and autonomous or semi-autonomous control, as human access is often limited or impossible. Additionally, systems must incorporate redundant safety features, reliable communication, and fault-tolerant designs to ensure continued operation in unpredictable and potentially life-threatening conditions, setting them apart from conventional robotic technologies. The high-energy reactions inside the reactor create intense heat, radiation, and strong magnetic fields, making direct human access for inspection and repairs impractical and dangerous. The complexity of these systems requires highly precise and timely maintenance to prevent failures, but the harsh conditions severely limit human intervention. Human intervention is frequently necessary in areas like diagnostics, repair, and system optimization, which increases operational costs and introduces potential risks of error. These interventions slow down the operational cycles and prevent the plants from running continuously, reducing overall efficiency. Additionally, the scale and complexity of future large, commercial fusion plants would demand continuous, proactive maintenance to ensure efficient and safe operations. To address these challenges, autonomous planning will be essential, enabling remote inspections and automated repairs, minimizing the need for human involvement in hazardous areas.

In future fusion power plants, autonomous planning (

Table 3. Summary of autonomous planning and execution in nuclear operations.

Aspect	Future Fusion Plant Maintenance	Fission Decommissioning
Initial Process	Autonomous inspection to detect system anomalies	Autonomous mapping of the environment without focusing on anomalies
Decision Generation	Decision generator creates maintenance strategies based on anomaly data and knowledge base	Decision generator formulates a decommissioning plan based on environment data, safety protocols, and regulations
Procedure Planning	Procedure generator sequences maintenance actions for human approval and robotic execution	Procedure generator sequences decommissioning steps aligned with safety and decision-making recommendations
Execution Approach	Remote handling robots perform approved tasks with human supervision in the loop	Robots carry out dismantling and decommissioning activities following predefined safe procedures
Autonomous Planning Role	Predict maintenance needs, optimize scheduling, propose solutions, reduce human intervention, and improve operational efficiency	Generate safe, efficient decommissioning strategies tailored to plant conditions, with minimal human exposure
Impact on Operations	Continuous, proactive maintenance, reduced downtime, enhanced safety, and optimized resource use	Safe, efficient decommissioning, regulatory compliance, and minimal human intervention in hazardous zones

) is essential for reducing human intervention and optimizing operational efficiency. Future fusion plants should prioritize automation to enhance operational speed and reliability, particularly in commercially viable applications. A key strategy could involve automating the process of remote inspection, where data are collected and analyzed autonomously. By integrating advanced automation technologies into the planning process, autonomous planning systems could analyze vast amounts of data to anticipate maintenance needs, identify potential issues, and optimize plant performance. These systems could autonomously generate maintenance schedules, prioritize tasks, and develop contingency plans to minimize downtime and ensure continuous operation. Upon identifying issues, potential solutions could be generated and proposed to human operators for approval. Once approved, these solutions would be executed using remote handling components, minimizing the need for human intervention. This approach would aim to streamline operations, reduce reliance on human inspection, and optimize the efficiency of fusion power plant operations.

Figure 7a describes the anticipated procedure of future plant maintenance to the best of our understanding, which could begin with an autonomous inspection to evaluate the system and identify any potential anomalies. If anomalies are not detected during the inspection, the process would continue with regular monitoring. However, if anomalies are found, a decision generator would be used to generate a suitable maintenance solution. This decision generator leverages a knowledge base to assess the anomalies and determine the most appropriate maintenance strategy based on predefined rules and relationships. Once a maintenance solution is determined, a procedure generator could provide a detailed sequence of actions required to accomplish the necessary maintenance and propose it to the human operator for approval. Having the human operator in the loop would allow the operators to oversee the activities and validate the approach. These sequences would carefully be designed to ensure the maintenance is carried out efficiently and effectively, minimizing downtime and restoring the system to optimal functionality. The whole procedure would repeat until full coverage of the vessel has been inspected.

Figure 7b details the anticipated procedure of future fission decommissioning to the best of our understanding, which could start with an autonomous inspection that could map the entire environment to gather detailed information about the system’s current state and structure. This mapping process would provide a comprehensive overview of the environment, including the location of various components and potential hazards, without focusing on anomalies. Based on the collected data, a decision generator could then be used to create a decommissioning solution tailored to the specific environment. This solution would consider various factors such as safety protocols, regulatory compliance, and optimal resource usage. Once a decommissioning solution is established, a procedure generator could provide a precise sequence of actions to guide the decommissioning process. This sequence would ensure that the decommissioning is carried out safely, efficiently, and in accordance with the decision generator’s recommendations, ultimately leading to the safe shutdown and dismantling of the fission system. This process would continue to repeat until the entire vessel has been inspected.

Autonomous planning could revolutionize nuclear operations in the future, offering innovative solutions to enhance efficiency and safety. By harnessing advanced automation technologies, autonomous planning systems could analyze complex datasets from various sources within nuclear operations, including sensors, diagnostics, and historical maintenance records. These systems could then autonomously identify trends, predict equipment degradation, and schedule proactive maintenance tasks to prevent unplanned downtime. Additionally, autonomous planning could optimize energy usage by dynamically adjusting operational parameters based on real-time demand and performance data. Furthermore, these systems could facilitate collaboration among different teams by providing centralized access to relevant information and recommending coordinated actions. Ultimately, autonomous planning has the potential to transform nuclear operations, ensuring reliable and cost-effective fusion research while minimizing human intervention.

In the future, the implementation of autonomous planning could profoundly impact operations at ITER, particularly in the realms of remote inspection and remote handling. Advanced automation technologies could empower robots to autonomously plan and execute inspection tasks throughout the ITER facility. These robots could analyze sensor data and historical inspection records to identify areas requiring assessment, plan optimal inspection routes, and adapt their strategies in real time based on environmental conditions. Additionally, autonomous planning could enhance remote handling operations by enabling robots to autonomously generate maintenance schedules, prioritize tasks, and coordinate with human operators. These robots could analyze data from remote sensors and diagnostics to anticipate maintenance needs, autonomously propose solutions, and execute tasks using remote handling equipment. By integrating autonomous planning into remote inspection and handling processes, ITER could enhance operational efficiency, optimize resource utilization, and minimize downtime, ultimately accelerating progress towards the development of practical fusion energy. Overall, autonomous planning will have the potential to enhance the efficiency, reliability, and safety of ITER’s operations, paving the way for the successful realization of fusion energy on a commercial scale (

Table 4. Robots mentioned in nuclear robotic remote operations.

Robot Name	Company/Institution	Model/Type
TARM	RACE (UKAEA)	Telescopic Articulated Remote Mast
Articulated Inspection Arm	CEA-LIST	Remote Inspection Robot
Spot	Boston Dynamics	Quadruped Mobile Robot
Spider	University of Manchester	Mobile Climbing Robot
AVEXIS	University of Manchester	Aquatic Inspection Vehicle
MASCOT	UKAEA (JET)	Bilateral Force Feedback Telemanipulator
MASCOT 6	RACE (UKAEA)	Upgraded Telemanipulator
Telbot	UKAEA	Telemanipulator

).

3. Object-Oriented Assembly and Disassembly Operations

3.1. Hardware Design and Creation

Hardware design and creation (Figure 8) provide a tangible and practical solution to physically manipulate components with precision. Grippers [42], designed to mimic the dexterity and control of a human hand, offer several advantages in these tasks. They can handle delicate or complex parts with care, ensuring accurate placement and reducing the risk of damage. Advanced grippers [43] equipped with sensors and adaptive control systems can dynamically adjust their grip based on the properties of the objects, enhancing flexibility and efficiency. There are also notable disadvantages in using hardware design and creation for object-oriented assembly and disassembly operations. Designing and building specialized grippers can be costly and time-consuming, requiring significant engineering expertise and resources. Additionally, while grippers can be highly effective for specific tasks, they may lack the versatility needed to handle a wide variety of objects and assembly scenarios without frequent reconfiguration or customization. Furthermore, integrating grippers into automated systems requires precise coordination and control, which can add complexity to the overall assembly process. Despite these challenges, the use of grippers in object-oriented assembly and disassembly operations remains a crucial component for achieving high precision and efficiency in automated manufacturing and assembly lines.

The development of multi-functional grippers for assembly and disassembly operations has garnered substantial attention in recent robotics research. For instance, a novel gripper concept (Figure 9a) was proposed in [44] for general grasping and in-hand dexterous manipulation. This gripper was able to perform bimanual manipulations using a single arm, thereby simplifying the robot structure and enabling a dual-arm system to utilize the second arm for tasks more meaningful than merely holding an object in place. In [45], a multi-functional robotic gripper equipped with a set of actions required for the disassembly of electromechanical devices was proposed. The system enabled 7-DoF manipulation and had the ability to reposition objects in hand and perform tasks which usually required bimanual systems. The proposed design could execute various disassembly actions rather than a disassembly line with multiple single robots with specific end effectors, thereby improving flexibility concerning disassembly type and order. A versatile manufacturing system was developed in [46], integrating CAD-based localization, compliance control, and a multi-functional gripper for swift and efficient programming of assembly tasks. Featuring an innovative gripper design with a parallel jaw element and a rotating module, the system was capable of addressing different types of assembly tasks without the need for retooling. However, the complexity of designing and manufacturing these grippers can lead to higher initial costs and potential maintenance challenges. These innovations collectively underscore the trend towards creating more flexible, efficient, and intelligent grippers capable of seamlessly transitioning between different assembly and disassembly tasks, enhancing productivity and reducing the need for multiple specialized tools.

The creation of low-cost grippers for assembly and disassembly tasks has become a focal point in recent robotics research, driven by the need to enhance efficiency while minimizing expenses. Traditional disassembly systems often involve high costs due to labor and complex mechanisms, which limit their accessibility for enterprises. Research efforts have increasingly focused on developing cost-effective alternatives that do not compromise performance. In [47], an innovative disassembly tool was introduced to facilitate the automation of disassembly processes. A cost-efficient screwnail (Figure 9b) was utilized as an end effector for the disassembly tool. A self-connection could be established during the screwnail indentation process, which enabled the transmission of forces and torques necessary for various dismantling operations. A cost-effective reconfigurable gripper was introduced in [48] to manage cylindrical and prismatic components within the dimensional ranges of electromechanical products. By employing these reconfigurable grippers for holding and manipulating components, the necessity for a series of dedicated stations with specialized grippers tailored for each component was eliminated, thereby facilitating flexible assembly and disassembly of electromechanical products. The trade-off of low-cost grippers may include potential limitations in durability and versatility, and they might lack the advanced features necessary for more complex or precise tasks. Balancing cost and functionality is essential to ensure that low-cost grippers can effectively enhance productivity and streamline operations in diverse assembly and disassembly applications.

3.2. Traditional Algorithms

Traditional algorithms encompass a wide range of computational techniques designed to solve complex problems efficiently and effectively. Control methods [49] provide real-time feedback, ensuring precise task execution and adjustment based on dynamic conditions. Optimal algorithms [50] focus on finding the best possible solution under given constraints, guaranteeing efficiency and minimizing resource use. Graphical methods [51] visually represent tasks, aiding in intuitive understanding and planning. Fuzzy logic [52] introduces flexibility by handling uncertainties and imprecise information, making it suitable for real-world scenarios where exact data may not always be available. Metaheuristic algorithms [53], such as genetic algorithms [54] and evolutionary algorithms [55], explore a wide solution space to find near-optimal solutions for problems that are too complex for other traditional methods. Geometric computation [56] deals with the spatial aspects of tasks, guaranteeing that components fit together correctly and optimizing the use of space. These traditional algorithms, with their diverse applications, continue to be foundational tools in advancing technology and solving real-world problems.

3.2.1. Control Algorithms

Control algorithms [57] offer a structured and systematic approach to managing complex tasks. These algorithms, including deterministic and feedback-based methods, provide precise control over the assembly and disassembly processes [58]. Advantages of control algorithms include their reliability and predictability, which are crucial for maintaining consistency and quality in repetitive tasks. They are well suited for environments with well-defined parameters and minimal variability. However, their rigidity can be a significant disadvantage in dynamic and uncertain environments where flexibility and adaptability are required. Control algorithms may struggle to handle unexpected changes or complexities in real time, leading to inefficiencies and potential errors [59]. Additionally, implementing these algorithms can be computationally intensive, requiring substantial resources and fine-tuning to achieve optimal performance. Despite these drawbacks, control algorithms remain a cornerstone in automated assembly and disassembly, providing a solid foundation for integrating more advanced, adaptive methods.

The application of control algorithms in assembly and disassembly operations has been extensively studied to enhance automation efficiency and robustness. Stochastic control deals with systems influenced by randomness and uncertainty. A decentralized, scalable approach to assembling a collection of heterogeneous parts into various products was developed in [60], which generated the strategy as a problem of selecting assembly and disassembly rates. These rates were translated into probabilities that defined stochastic control policies for individual robots, which in turn produced the desired collective behavior. Stochastic control can manage uncertainty and randomness, optimizing performance in unpredictable environments, though they can be computationally intensive. Open-loop control operates without feedback, relying on predetermined inputs to achieve desired outcomes, which is simple and cost-effective. Nevertheless, it lacks responsiveness to disturbances. In [61], an innovative modular design based on a quadrotor platform that utilized a lightweight passive mechanism for midair docking and undocking was introduced. For the undocking strategy, rapid torque generation in open-loop control was proposed, ensuring that the aerial modules are capable of performing inflight self-reconfiguration. Closed-loop control (Algorithm 1) continuously adjusts based on real-time feedback, enhancing accuracy and stability, albeit at a higher complexity and cost. In [62], the authors addressed the challenge of self-assembling magnetic modular cubes into specified 2D polyomino shapes utilizing magnetic fields. A closed-loop control method for self-assembling the magnetic modular cubes into polyomino shapes was introduced, employing computer vision-based feedback with re-planning. The proposed closed-loop control method enhanced the success rate of forming user-specified polyominoes in comparison with the open-loop baseline. Adaptive control further refines closed-loop systems by dynamically adjusting parameters in response to changing conditions, offering superior flexibility but requiring sophisticated algorithms. A concept for generating flexible control sequences for a semi-automated disassembly system was introduced in [63]. Achieving this flexibility required module-based adaptive control. Control strategies can lead to more flexible and resilient assembly and disassembly systems, reducing dependency on predefined conditions.

Algorithm 1: An example of closed-loop control for assembly/disassembly

3.2.2. Optimal Algorithms

Optimal algorithms [64,65] provide a strategic approach to achieving the most efficient and effective sequences of tasks. These algorithms aim to minimize resource consumption, such as time and materials, by determining the best possible order and methods for assembly and disassembly operations. The primary advantage of optimal algorithms is their ability to produce highly efficient solutions that can significantly reduce operational costs and improve productivity [66]. They are particularly useful in environments where precision and efficiency are paramount. However, the major disadvantage of optimal algorithms is their computational complexity [67]. Finding the optimal solution often requires extensive calculations and can be time-consuming, especially for large-scale or highly complex tasks. This can make them impractical for real-time applications or situations with high variability. Additionally, developing and implementing these algorithms can be challenging, requiring expertise in mathematical modeling and optimization techniques. Despite these drawbacks, optimal algorithms are invaluable for maximizing efficiency and performance in object-oriented assembly and disassembly operations.

Optimal algorithms are designed to handle complex information and streamline decision-making processes, breaking down intricate tasks into manageable steps. An intelligent virtual assembly system was introduced in [68], where an optimal assembly algorithm was utilized to facilitate haptic interactions during virtual assembly operations. The optimized haptic path and sequence guidance improved the efficiency of virtual assembly tasks. In [69], a planning system was developed to address long-horizon multi-robot construction assembly challenges, integrating several innovative components. By combining optimization techniques to address manipulation constraints with a sampling-based path planner, cooperative multi-robot manipulation with indeterminate arrival times was achieved. In [70], the issue of sequential task assignment and collision-free routing for large teams of robots in scenarios with inter-task precedence constraints was investigated. A hierarchical algorithm was proposed to compute makespan-optimal solutions for this problem. The output of this algorithm could be utilized to obtain collision-free trajectories for differential-drive robots. An approach capable of automatically and reliably completing high-accuracy robotic peg-in-hole assembly (Figure 10) was introduced in [71]. A trajectory optimization strategy was proposed, incorporating admittance control, exception-handling techniques, and multiple trajectory iteration methods. The high-accuracy assembly experiments yielded positive results, confirming the effectiveness of the proposed peg-in-hole robotic assembly approach. However, optimal algorithms may require sophisticated programming and computational resources. By optimizing sequences and actions, these algorithms enhance operational efficiency, reduce costs, and improve overall system performance.

Optimal algorithms are revolutionizing automation by significantly reducing the time required for these processes. In [72], the balancing of the robotic disassembly line problem was examined to develop efficient solution techniques. A heuristic algorithm based on ant colony optimization was introduced to find solutions for particularly large-scale test problems, given the problem’s complexity. The computational results indicated that the proposed mathematical model and algorithms were promising for both small- and large-scale test problems. Taguchi’s orthogonal arrays were introduced in [73] to evaluate the robustness of the Simulated Annealing algorithm. The proposed model also demonstrated robustness, delivering reliable results under different conditions of the disassembly platform in addition to its efficiency in producing optimal solutions. This approach was able to integrate the preferences of decision-makers and provide robust and reliable solutions to solve disassembly sequencing problems. The authors in [74] addressed a novel issue involving robotic disassembly sequence planning for end-of-life products and developed a corresponding multi-objective model to minimize energy consumption and makespan. A discrete artificial bee colony algorithm [75] was proposed to derive high-quality disassembly schemes. The proposed method effectively reduced the makespan and energy consumption. In [76], a new AND/OR-graph-based disassembly sequence planning approach was developed by considering fluctuating disassembly operational costs as well as component quality. The proposed method generated the optimal solution for the fuzzy disassembly sequence optimization problem. The results could be used to guide decision-makers and provide a new analytical approach to determining the best disassembly sequence. Implementing optimal algorithms can be computationally intense and complex, which can be a drawback. Nevertheless, the ability of optimal algorithms to swiftly and accurately determine the best paths and actions for assembly and disassembly tasks can handle a wide range of variables and uncertainties, ensuring robust and reliable outcomes.

3.2.3. Graphical Methods

Graphical methods [77] provide an intuitive and visual approach to managing complex tasks. These methods involve the use of diagrams, flowcharts, and other visual tools to represent the components, relationships, and sequences involved in assembly and disassembly processes [78]. The primary advantage of graphical methods is their ability to simplify complex information, making it easier to understand and communicate. This visual clarity can enhance decision-making, improve planning accuracy, and facilitate collaboration among team members [79]. Additionally, graphical methods can help identify potential issues and optimize workflows by visually mapping out each step of the process. However, there are also disadvantages to this approach. Graphical representations can become overly complicated and difficult to manage when dealing with highly intricate systems or large-scale tasks [80]. They may also require specialized software and skills to create and interpret effectively. Moreover, while graphical methods are excellent for initial planning and communication, they may not provide the level of precision needed for detailed execution and may need to be complemented with other analytical techniques. Despite these challenges, graphical methods remain a valuable tool for enhancing understanding and efficiency in object-oriented assembly and disassembly operations.

The use of graphical methods to handle uncertainties and different probabilities in assembly and disassembly operations offers a robust approach to optimizing complex processes. The objective of [81] was to structure the disassembly of reusable equipment using a model to develop structured and optimized strategies. To organize maintenance tasks collectively and reduce uncertainties, aspects of graphical methods and Petri nets [82] were considered in disassembly applications to define the model. One of the drawbacks of employing optimization methods to determine disassembly sequences is that the input parameters for the optimization algorithms are often unavailable or estimated with high uncertainty. To tackle this issue, an application of immersive computing technology was proposed in [83] to assist in the evaluation and training of product disassembly sequences using interactive graph visualizations. The drawbacks of using graphical methods to deal with uncertainties include the need for precise and up-to-date data, the potential complexity in constructing and interpreting the models, and the risk of oversimplifying complex interactions. In [84], a graphical method was introduced to automatically generate all possible sequences for disassembling a mechanical system, aimed at enhancing maintenance and recycling activities. This method significantly reduced the number of disassembly operations required for complex systems while maintaining generality, thereby enabling the precise calculation of all disassembly sequences. In [85], a stability graph cut-set method was applied to generate the best disassembly sequences. To address uncertainties, the best assembly sequences were generated by taking directional changes as a fitness function. The benefits of increased adaptability and robust planning make graphical methods essential tools for managing uncertainties in assembly and disassembly operation processes.

3.2.4. Fuzzy Logic

Fuzzy logic [86] introduces a flexible and adaptive approach to handling uncertainties and imprecise information. Fuzzy logic systems operate on the principle of partial truth, where truth values range between completely true and completely false [87]. This approach allows for more nuanced decision-making in complex and uncertain environments. The primary advantage of fuzzy logic is its ability to model and manage ambiguity, making it well suited for tasks where precise data are unavailable or where human-like reasoning is beneficial [88,89]. It enhances robustness and flexibility in the face of variability and unexpected changes. However, fuzzy logic also has its disadvantages. Designing and tuning a fuzzy logic system can be challenging, as it requires expertise in defining appropriate membership functions and rules [90]. Additionally, while fuzzy logic can simplify decision-making processes, it may introduce complexity in the interpretation and analysis of results. Moreover, its reliance on heuristic rule sets can lead to less predictable outcomes compared to more deterministic methods. Despite these challenges, fuzzy logic remains a powerful tool for enhancing adaptability and resilience in object-oriented assembly and disassembly operations.

The application of fuzzy logic to obtain a dynamic decision-making process for assembly and disassembly operations offers a sophisticated approach to managing the complexities and uncertainties inherent in these operations. A hybrid fuzzy logic–genetic algorithm approach was proposed in [90] to plan automatic disassembly and assembly sequences of products. A fuzzy controller was developed to dynamically adjust the mutation probability and crossover rate during the algorithm’s execution. To address uncertainty in a dynamic decision-making process, a fuzzy reasoning Petri net model was developed in [91] to represent the pertinent decision-making rules in the disassembly process. Utilizing the proposed fuzzy reasoning algorithm based on this model, multicriteria disassembly rules were considered in parallel to enable automatic and rapid decision-making. However, the implementation of fuzzy logic requires careful tuning and validation to ensure accuracy and reliability. In [92], a fuzzy attributed Petri net model was developed to mathematically represent the uncertainty in disassembly caused by significant human intervention. An algorithm based on this model was further proposed for optimal disassembly planning, aiming to enhance the technique’s applicability to real industrial settings. The ability of fuzzy logic to enhance adaptability and robustness makes it a valuable tool for improving the efficiency and effectiveness of assembly and disassembly operation processes.

3.2.5. Metaheuristic Algorithms

Metaheuristic algorithms [93] offer flexible and powerful approaches to solving complex optimization problems. Metaheuristics, such as genetic algorithms [94], and particle swarm optimization, are designed to explore large search spaces efficiently and find near-optimal solutions within a reasonable time frame. The primary advantage of these algorithms is their ability to handle complex, nonlinear, and multi-modal problems where other traditional methods may fail or be impractical [95]. They are particularly effective in adapting to changing conditions and can be applied to a wide range of scenarios without requiring precise mathematical models. However, metaheuristic algorithms also have some disadvantages. They are often not guaranteed to find the absolute optimal solution and may converge on local optima instead. Despite these challenges, metaheuristic algorithms are highly valued for their robustness and adaptability, making them suitable for the dynamic and complex nature of object-oriented assembly and disassembly operations.

The use of metaheuristic algorithms in assembly and disassembly operations represents a cutting-edge approach to quickly adapt to changes and improve operational efficiency. In [96], an assembly sequence and path planner/replanner were introduced to address the assembly sequence and path planning problem. The proposed approach featured a straightforward greedy heuristic for sequence planning and a sampling-based stochastic component for path planning. This method had the advantage of accounting for obstacles within the workspace, enabling the planning of both translational and rotational movements for parts, and managing non-monotone assembly sequence plans. However, implementing metaheuristic algorithms can be computationally intensive and may require fine-tuning to balance exploration and exploitation effectively [97]. To enhance disassembly efficiency, a novel cooperative disassembly sequence and task planning method was developed in [98], utilizing a genetic algorithm. A mathematical model was formulated based on the disassembly hybrid graph model. The chromosome evolution rules were restructured to achieve optimal multiplayer task planning and cooperative disassembly sequences. The authors in [99] explored flexible robotic disassembly sequencing in the context of failed automation operations and developed online recovery mechanisms through the incorporation of backup actions. A bi-objective optimization model for disassembly sequence planning was formulated using a dual-selection multi-objective evolutionary algorithm [100]. These backup actions enabled efficient recovery from failures and potentially enhanced the robustness of the robotic disassembly process. Due to factors such as complex geometric computations, obstacles, orientation, and the initial or final position of parts, path planning becomes a critical factor influencing the efficiency of the maintenance process. To tackle this issue, an intelligent assembly planner was proposed in [101], utilizing a combined approach based on genetic optimization techniques and potential field methods for maintenance assembly and disassembly path planning. The significant improvements in adaptability and efficiency underscore the value of metaheuristic algorithms in advancing assembly and disassembly operation processes.

The application of metaheuristic algorithms in assembly and disassembly operations is an innovative approach aimed at minimizing both time and cost. In [102], a metaheuristic algorithm was proposed to determine the optimal disassembly sequence. Operations were clustered into cells based on the resources they required, with the objective of minimizing machine acquisition costs. The goal was to group operations utilizing similar equipment to ensure high utilization levels of such machinery. Nevertheless, the implementation of metaheuristic algorithms may involve substantial computational resources and require careful calibration to achieve the desired balance between solution quality and computational effort. To minimize the anticipated total cost of the disassembly service and the expected total makespan, a mathematical model accounting for the precedence relationships of disassembly tasks and the uncertain nature of the disassembly process was proposed in [103]. A multi-objective genetic algorithm, based on the non-dominated sorting genetic algorithm II [104], was designed to solve this mathematical model. The potential for substantial improvements in efficiency and cost-effectiveness makes metaheuristic algorithms a valuable tool in the realm of assembly and disassembly operations.

3.2.6. Geometric Computation

Geometric computation [105] involves leveraging mathematical and computational techniques to manage spatial relationships and configurations. This approach is particularly advantageous for tasks requiring precise positioning, orientation, and alignment of components [106]. The primary advantage of geometric computation is its accuracy in handling spatial data, which ensures that assemblies are constructed and disassembled with exactitude. It allows for detailed modeling of geometric properties and spatial constraints, facilitating the optimization of space utilization and the detection of potential interferences or collisions [107]. However, there are notable disadvantages as well. Geometric computation can be computationally intensive, requiring significant processing power, especially for complex or large-scale assemblies [108]. This can lead to longer computation times and increased resource requirements. Additionally, implementing geometric computation often demands specialized software and expertise in computational geometry, which can add to the complexity and cost of the process [109]. Furthermore, although geometric methods excel in handling spatial data, they may not adequately address non-geometric factors such as material properties or dynamic interactions between components. Despite these challenges, geometric computation remains a crucial tool for achieving high precision and efficiency in object-oriented assembly and disassembly operations.

The use of geometric computation in assembly and disassembly operations introduces time-saving benefits to automated systems. A method that accounted for the collision relationships and geometric contact among components was introduced in [110] to generate the disassembly geometry contacting graph. This proposed approach removed all components unrelated to the target, thereby reducing the search time. Most assembly sequence generation methods lack the flexibility to accommodate assembly tasks performed concurrently by the operator and robot. To address this issue, an algorithm for identifying assembly task precedence was presented in [111]. This approach allowed for parallel task execution, unhindered by linear assembly sequences, thereby saving operation time. However, implementing geometric computation requires sophisticated algorithms and computational resources to manage the detailed geometric data effectively. In [112], a geometric reasoning engine was developed to identify the optimal disassembly plan. To expedite the optimization process, a best-first search algorithm was used to quickly determine a disassembly sequence solution. A caching technique was devised to reuse feasible disassembly operations computed in previous steps, thereby reducing computational time. A decomposition approach that segregated the macro-level planning problem was developed in [113], which handles task sequencing and resource allocation, from the micro-level validation involving detailed technological and geometrical models. The geometric inference methods employed adaptations of established techniques to generate robust disjunctive constraints on the assembly plan. This proposed methodology ensured the feasibility and saved time in the assembly plans of detailed models. The benefits of reduced processing times make geometric computation an invaluable asset in modern manufacturing and industrial automation.

3.3. Neural Networks

Neural networks (NNs) introduce advanced computational techniques that enhance automation and decision-making capabilities. NNs [114], particularly deep learning models [115], can handle complex patterns and relationships within large datasets. Feature extraction methods [116] enable these networks to identify and focus on relevant aspects of tasks, improving accuracy and efficiency. Deep learning models [117] can learn intricate representations and make precise predictions, which are valuable for optimizing task sequences and detecting anomalies. Reinforcement learning [118] further enhances these processes by allowing NNs to learn from interactions with the environment, continuously improving their performance through trial and error. NNs offer significant potential for enhancing the efficiency, adaptability, and intelligence of various applications.

3.3.1. Feature Extraction

Feature extraction [119] involves identifying and isolating relevant characteristics from complex data to enhance task performance. This method focuses on extracting essential information, such as shapes, patterns, and spatial relationships, which are critical for accurately assembling and disassembling components [120]. The primary advantage of feature extraction is its ability to reduce data complexity, making it easier to analyze and process information. By focusing on key features, the system can achieve higher accuracy and efficiency, particularly in tasks that require precise identification and manipulation of parts. Additionally, feature extraction facilitates the integration of advanced algorithms, such as machine learning and computer vision, improving the overall automation and intelligence of the process. However, there are also disadvantages to consider. Feature extraction can be challenging and time-consuming, requiring significant expertise to identify the most relevant features and ensure they are accurately captured [121]. The quality of the extracted features directly impacts the performance of subsequent algorithms, meaning any inaccuracies or omissions can lead to suboptimal results. Moreover, feature extraction methods must be carefully tailored to the specific requirements of each task, which can limit their generalizability and require continuous adjustments. Despite these challenges, feature extraction remains a vital tool for enhancing the precision and efficiency of object-oriented assembly and disassembly operations.

The application of feature extraction in assembly and disassembly operations significantly enhances accuracy and efficiency in automated systems. An unsupervised learning framework was introduced in [122] for video-based human activity recognition. A new model for unsupervised feature extraction was devised, which merged the capability of convolutional neural networks (CNNs) to analyze spatial image data with the strengths of recurrent neural networks for processing temporal data. This framework demonstrated an average recognition accuracy surpassing that of existing methods. In [123], a computer vision framework was proposed for detecting screws and recommending appropriate tools for disassembly. This framework employed a modified YOLOv4 algorithm to identify screw targets and implemented a screw image extraction mechanism based on the predicted position coordinates. The alterations to the YOLOv4 algorithm enhanced screw detection accuracy. In [124], a novel framework for disassembly sequence planning of industrial components was introduced. This framework utilized a convolutional neural network to differentiate between fasteners and non-fasteners. The extracted information was used to prioritize fasteners within the disassembly process loop, thereby enhancing the accuracy and efficiency of disassembly planning. However, implementing feature extraction requires computational resources to effectively process and analyze component data. The substantial improvements in accuracy and efficiency make feature extraction a vital tool in advancing automation in manufacturing and industrial settings.

3.3.2. Deep Learning

Deep learning (Algorithm 2) offers a sophisticated approach to managing complex and dynamic tasks. Deep learning models, particularly NNs with multiple layers, can automatically learn and model intricate patterns and relationships within large datasets [125]. This capability allows for highly accurate predictions and optimizations in task sequences, improving the efficiency and precision of assembly and disassembly processes. One significant advantage of deep learning is its ability to handle unstructured data, such as images and sensor data, making it particularly effective when integrated with computer vision for identifying and manipulating components [126]. Additionally, deep learning models can adapt to new tasks and variations without explicit programming, enhancing flexibility and scalability. However, there are also disadvantages to using deep learning in this context. Training deep learning models requires substantial computational resources and extensive datasets, which can be time-consuming and costly to acquire [127]. The complexity of these models can also lead to longer development times and increased difficulty in debugging and fine-tuning. The decision-making process is often opaque, making it difficult to interpret and understand the model’s reasoning. Despite these challenges, deep learning remains a powerful tool for advancing the capabilities of object-oriented assembly and disassembly operations, offering significant improvements in automation, adaptability, and performance.

Algorithm 2: An example of deep learning for assembly/disassembly

The application of deep learning in assembly and disassembly operations is revolutionizing automation by significantly increasing precision and efficiency. In [128], experimental data collection was integrated with deep learning models to achieve precise disassembly task recognition. Additionally, a sequence-based correction algorithm was proposed to enhance the precision of the entire process. In [129], a graph transformer-based framework for addressing the assembly sequence planning problem was introduced and validated using a self-compiled assembly sequence planning database. Utilizing this bespoke dataset, a heterogeneous graph transformer model was proposed to discern the underlying rules of assembly planning. The similarity between the predicted sequences and the ground truth sequences exhibited a moderate correlation. The authors in [130] presented a methodology for optimizing the balance between input resolution and mini-batch size to achieve peak performance in cross-recessed screw detection while maximizing graphics processing unit utilization. Tiny-YOLO v2 was utilized to address the specific task of detecting cross-recessed screws. This model demonstrated high average precision across all test datasets. The authors in [131] explored a deep learning system that combines CNNs with long short-term memory (LSTM) networks for vision-based signal processing to accurately predict human motion. This approach not only eliminated the need for complex feature extraction but also facilitated seamless interaction between humans and robots without the need for wearable devices, which can be cumbersome. The proposed method surpassed three other optimization algorithms in prediction accuracy for desktop computer disassembly. While the implementation of deep learning requires substantial computational resources and expertise in algorithm development, the resulting advancements in precision and efficiency make it an indispensable tool for modern industrial automation.

3.3.3. Reinforcement Learning

Reinforcement learning [132] introduces a dynamic approach where algorithms learn optimal strategies through trial-and-error interactions with the environment. In this method, an agent receives feedback in the form of rewards or penalties based on its actions, progressively improving its performance over time [133]. The primary advantage of reinforcement learning is its ability to handle complex and variable environments, adapting to new conditions without explicit programming [134]. This adaptability makes reinforcement learning particularly valuable for tasks requiring continuous learning and real-time decision-making, such as adjusting assembly sequences in response to changing component conditions or unexpected obstacles. However, there are also disadvantages to consider. Training reinforcement learning models can be resource-intensive, requiring substantial computational power and time, especially for complex tasks with large state and action spaces [135]. The learning process may involve a significant amount of trial and error, which can be inefficient and costly in practical applications [136]. Additionally, designing effective reward functions that accurately reflect the desired outcomes can be challenging and may require extensive experimentation. Despite these challenges, reinforcement learning offers significant potential for enhancing the automation, flexibility, and efficiency of object-oriented assembly and disassembly operations, making it a promising tool for advanced industrial applications.

The use of reinforcement learning in assembly and disassembly operations is paving the way for significant reductions in task complexity and substantial increases in productivity. Due to the inherent complexity that arises from constructing diverse blueprints using a set of blocks, the authors in [137] discovered that combining large-scale reinforcement learning with graph-based policies, without introducing additional complexity, effectively trains agents. These agents not only generalize to intricate unseen blueprints in a zero-shot manner but also function in a reset-free environment without prior training for such conditions. In order to enhance the productivity of robotic work cells, a computational algorithm was proposed in [138] for identifying an efficient assembly sequence and assigning tasks to robotic hands using reinforcement learning. The proposed methods aimed to reduce or eliminate the workload involved in robot teaching while also shortening product assembly time by providing an efficient work plan. In [139], a novel method was presented for autonomously learning an assembly task beginning with initial disassembly. Hierarchical reinforcement learning was proposed for the disassembly policy, which broke down the learning process into high-level decision-making and a lower-level compliant controller. Utilizing information gathered during disassembly, the autonomous learning process can be effectively achieved and simplified. Despite the need for extensive training data and computational resources, the ability of reinforcement learning to handle complex, dynamic environments makes it a powerful tool for advancing efficiency in industrial automation.

The application of reinforcement learning in assembly and disassembly operations offers unprecedented adaptability to new tasks, significantly enhancing the flexibility and efficiency of automated systems. A platform enabling robots to learn disassembly using reinforcement learning was introduced in [140]. The efficacy of the platform was illustrated through a robot that learned to remove a bolt along a door-chain groove. Skills acquired by a low-precision robot could be transferred to a robot with high precision, demonstrating enhanced performance and stability post-transfer. A combination of reinforcement learning and planning was proposed in [141] for assembling structures. Developing an analytical model for a controller to manage various contact interactions was exceedingly difficult, making reinforcement learning a more viable approach. The proposed framework supported component reuse, as well as the reassembly and disassembly of parts. In [142], a framework named Form2Fit was introduced for generalizable kit assembly. By disassembling kits through pick-and-place actions, this self-supervised system generated its training labels for assembly and then reversed action sequences. By treating the assembly task as a problem of shape-matching, the proposed method learned a robust matching function that was adaptable to various initial conditions, capable of managing new kit combinations, and able to generalize to unfamiliar kits and objects. Although implementing reinforcement learning algorithms requires substantial computational resources and extensive training, the resulting ability to swiftly adapt to new challenges makes reinforcement learning an invaluable tool for modern automation in manufacturing and other industries.

3.4. Human–Robot Collaboration

Human–robot collaboration [143] leverages the strengths of both humans and robots to achieve higher efficiency and precision. Methods such as human-assisted tasks [144], where robots learn from human demonstrations, enable robots to acquire skills and adapt to new tasks by observing human actions. Human–robot cooperation [145] involves seamless interaction between humans and robots, where tasks are dynamically shared based on the strengths of each partner. For instance, humans might handle complex decision-making and delicate manipulations, while robots perform repetitive or heavy tasks. Robot-assisted tasks [146] involve robots supporting human operators by generating solutions that consider human safety constraints or providing tools, positioning parts, and performing auxiliary actions for human operators. This collaboration ensures that humans are protected from hazardous conditions and physical strain, enhancing workplace safety and productivity. Human–robot collaboration holds significant promise for different applications, combining the best of human and robotic capabilities to achieve superior outcomes.

3.4.1. Human-Assisted Tasks

Human-assisted tasks [147], where humans assist robots, leverage human intuition and dexterity to train and guide robotic systems. In this approach, humans perform tasks while robots observe and learn from demonstrations. This method, often implemented through techniques like teaching or programming by demonstration, allows robots to acquire complex skills that would be challenging to program explicitly [148]. The primary advantage of this approach is the ability to rapidly transfer expert knowledge to robots, enabling them to handle intricate tasks with greater precision and adaptability. It allows for quick reprogramming and customization of robotic behavior to suit specific tasks or environments. However, there are also disadvantages to consider. The process of human-assisted training can be time-consuming and may require significant effort from human experts to ensure the robot learns accurately [149]. Additionally, the initial setup and calibration for effective human–robot interaction can be complex and resource-intensive. There is also the potential for variability in human performance, which can affect the consistency of the training data. Despite these challenges, human-assisted tasks provide a powerful means of enhancing robotic capabilities in object-oriented assembly and disassembly operations, combining human expertise with robotic efficiency and consistency.

The integration of human-assisted tasks (Figure 11) in robotic assembly and disassembly operations significantly reduces task complexity and enhances overall efficiency. Given that most current human prediction algorithms struggle with poor generalization and extensive training data needs, the authors in [150] modeled the human as a rational entity aiming to minimize unknown cost functions along its motion trajectory. They employed maximum entropy inverse reinforcement learning to infer underlying cost functions from human demonstrations. Since industrial robots used for assembling customized products required extensive reprogramming, an algorithm was proposed in [151] to mitigate programming complexity by autonomously identifying the fastest collision-free assembly plans. A digital twin of the robot employed a simulation gym to discover which assembly skills, programmed by demonstration, were physically free from environmental collisions. The authors in [152] introduced a robotic learning framework through multiple human demonstrations. By observing these demonstrations, the robot acquired the sequence of necessary actions for assembly without requiring pre-programming. Once the robot had assimilated the demonstrated task, it would be capable of performing the task collaboratively with a human. In [153], a novel teaching–learning collaboration model for collaborative robots to assist their human partners and learn from human demonstrations was presented. The robot utilized the maximum entropy inverse reinforcement learning algorithm to gain knowledge from human assembly demonstrations, updating its task-based knowledge with the optimal assembly solution. Although human-assisted tasks require effective communication and coordination between humans and robots, the resulting reduction in task complexity and improvement in operational flexibility highlights the value of human-assisted robotic systems in modern industrial applications.

3.4.2. Human–Robot Cooperation

Human–robot cooperation [154] involves humans and robots working together simultaneously to complete tasks, leveraging their complementary strengths. In this cooperative setup, humans handle complex, nuanced activities requiring cognitive skills, creativity, and fine motor control, while robots manage repetitive, precise, or physically demanding tasks. This synergy enhances overall productivity and efficiency, as humans can focus on tasks that require judgment and adaptability, while robots provide consistency, speed, and endurance. One of the primary advantages of human–robot cooperation is the ability to tackle complex assemblies that neither humans nor robots could efficiently handle alone [155], leading to improved product quality and reduced production time. However, there are also challenges associated with human–robot cooperation. Ensuring seamless and safe interaction between humans and robots requires sophisticated control systems, reliable communication protocols, and intuitive interfaces. This complexity can increase the initial setup time and cost. Additionally, there are significant safety concerns, as robots must be equipped with advanced sensors and algorithms to detect and respond to human presence, preventing accidents [156]. Despite these challenges, human–robot cooperation offers a promising approach to optimizing object-oriented assembly and disassembly operations, combining human ingenuity with robotic precision and strength.

The utilization of human–robot cooperation (Figure 12) in assembly and disassembly operations is a strategic approach to effectively managing uncertainties and enhancing operational efficiency. Given the high-level uncertainty and lack of reliability in robotic systems for complex disassembly tasks, a disassembly planning method based on human–robot collaboration was proposed in [157]. This method leveraged the flexibility and problem-solving abilities of humans in conjunction with the repeatability and precision of robots. To enhance process efficiency, components were prioritized according to the remanufacturability parameters. To mitigate the effects of uncertainties, the authors in [158] introduced an innovative robotic disassembly cell featuring a human operator and two collaborative robots. The operator and robots could safely collaborate on parallel, individual, or shared disassembly tasks within a common workspace. This approach facilitated the automation of complex tasks, potentially freeing humans from tedious labor. In [159], a hierarchical framework for task allocation in human–robot cooperative assembly was proposed. This meticulous design enabled the system to handle unpredictable events at the decision-making level without the need for explicit planning. These unpredictable events were common in partially known and dynamic environments. Aiming to address the challenge of close proximity collaboration with partial occlusions, a vision-based 3D hand–object pose estimation method was developed in [160] for human–robot collaborative disassembly. An explicit occlusion area prediction was incorporated as a regularization term during training. This enhancement increased the robustness of the model to partial occlusion uncertainties between the object and the hand. Although human–robot cooperation necessitates advanced communication protocols and synchronization mechanisms, the synergy achieved through human–robot cooperation is invaluable in addressing the uncertainties inherent in complex assembly and disassembly tasks.

Utilizing human–robot cooperation in assembly and disassembly operations presents a powerful strategy for saving time and enhancing productivity. In [161], a comprehensive disassembly sequence planning algorithm was developed within a human–robot collaboration framework, taking into account critical factors such as resource limitations and worker safety. The objective was to reduce the total disassembly time while adhering to resource constraints and ensuring safety. The proposed algorithm effectively planned and allocated disassembly tasks among robots, human operators, and collaborative human–robot efforts. Since concurrent human–robot disassembly operations could enhance productivity and reduce costs, the authors in [162] introduced an interactive disassembly planning approach leveraging human–robot collaboration. The proposed method devised optimal disassembly strategies by allocating tasks between humans and robots. Using relationship matrices and an industrial database, the approach estimated the total disassembly time for the generated plan, minimizing changes in tool usage and dismantling directions. A methodology for distributing tasks between humans and robots in assembly work through complexity-based task classification was proposed in [163]. The approach categorized tasks by their complexity in mounting, part feeding, handling, and human safety, distinguishing higher-complexity tasks from those of lower complexity. This structured method for task allocation in human–robot collaboration simplified the automation process and significantly reduced changeover and deployment times. Despite the need for sophisticated coordination and communication systems, the significant time savings and improved operational efficiency underscore the value of human–robot cooperation in modern industrial settings.

3.4.3. Robot-Assisted Tasks

Robot-assisted tasks [164] involve robots aiding humans by performing supporting roles, such as holding tools, positioning parts, or executing repetitive actions. This collaboration allows humans to focus on tasks requiring critical thinking, precision, and dexterity, while robots handle monotonous or physically strenuous activities. The primary advantage of this approach is the enhancement of human productivity and ergonomics, as robots can take over labor-intensive tasks, reducing physical strain and the risk of injury for human workers [165]. This leads to a safer and more comfortable working environment and increases overall efficiency and throughput in the assembly and disassembly processes. However, there are also disadvantages to consider. Integrating robots into human-centric workflows requires careful planning and coordination to ensure seamless interaction, which can be complex and costly [166]. The initial investment in robotic systems and their maintenance can be substantial. Additionally, there is a learning curve for human workers to adapt to working alongside robots, which may require additional training and adjustment periods. Ensuring the safety of human workers in close proximity to robots is also a critical concern, necessitating advanced safety features and protocols. Despite these challenges, robot-assisted tasks hold significant potential for optimizing object-oriented assembly and disassembly operations by enhancing human capabilities with robotic support.

Implementing robot-assisted tasks in assembly and disassembly processes is an innovative approach aimed at decreasing human workload and enhancing overall efficiency. An innovative collaborative architecture was proposed in [167] for human–robot assembly tasks. With the proposed framework, humans were capable of performing more dexterous tasks, while robots assisted in the assembly process by reducing the physical and cognitive workload of humans, thereby minimizing errors and reducing absenteeism. A framework for planning collaborative robot assembly tasks was presented in [168], taking into account the intentions of both the operator and the designer. Potential assembly plans were automatically generated and converted into a state graph reflecting the operator’s intentions. For natural collaboration, the robot must estimate the assembly intentions of the operator and act accordingly. In [169], an interactive user and planner framework for part disassembly or assembly tasks was proposed, featuring rapid motion planning and real-time guiding forces. An algorithm was developed to seek a collision-free path in real time based on the computed data. Upon identifying the path, haptic artificial guidance was provided to the user who was able to influence the planner by either adhering to or disregarding the path, thereby triggering automatic new path searches. In [170], a novel method was introduced for automating screw unfastening. Robots were programmed to execute a spiral search motion to align the tool with the screw. Decreasing the disassembly labor content through automation was crucial for enhancing the economic viability of remanufacturing. While robot-assisted tasks require effective integration and coordination between humans and robots, the resulting benefits in workload reduction and operational efficiency make this approach an invaluable asset in modern industrial environments.

Implementing robot-assisted tasks in assembly and disassembly processes offers a strategic advantage in saving time and boosting productivity. In [171], a robotic knowledge graph was introduced to assist individuals lacking the necessary expertise to perform disassembly tasks. A prototype system was developed to help users manage, analyze, and acquire disassembly knowledge by integrating graph-based knowledge representation. The proposed knowledge graph effectively reduced idle periods, human workload, and disassembly time, providing essential knowledge to both human operators and robots during the disassembly process. Since efficiently obtaining an optimal sequence typically posed a challenge due to high computational costs, a disassembly task planning algorithm was presented in [172], incorporating human behavior prediction and human–robot collaboration to solve this challenge. The cost function accounted for the efforts of both humans and robots, considering time spent on tasks as well as movement distance. A disassembly sequence planner was developed in [173], distributing tasks between humans and robots within a collaborative framework. The proposed planner identified the orientations and locations of items to be disassembled and generated the optimal disassembly sequence while adhering to disassembly rules. The robot successfully located and disassembled the components according to the optimal sequence, completing the task collaboratively with humans without breaching any constraints. Although the integration of such systems requires sophisticated coordination, the substantial time savings and enhanced productivity highlight the immense value of robot-assisted tasks in modern industrial applications.

3.5. Knowledge Representation

Knowledge representation [174] involves employing structured approaches to encode and process the necessary information efficiently. Ontology models [175] provide a hierarchical framework that captures the relationships and attributes of various components and tasks, facilitating clear communication and interoperability between different systems and tools. Domain-specific languages [176] offer tailored programming solutions, enhancing clarity and precision. The planning domain definition language (PDDL) [177] is used to define the parameters and constraints of tasks, enabling automated planners to generate and solve complex sequences efficiently. Tree-based methods, such as decision trees [178] and rapidly exploring random trees [179] (RRTs), are employed to explore possible actions and their outcomes systematically, aiding in the efficient navigation and decision-making process within the task space. Knowledge representation techniques are crucial for advancing automation, improving decision-making, and optimizing the efficiency of various applications.

3.5.1. Ontology Models

Using ontology models [180] to manage object-oriented assembly and disassembly operations involves creating a structured representation of the knowledge within this domain. Ontology models define the components, their properties, relationships, and the rules governing their interactions in a hierarchical and interconnected manner. This structured approach enhances interoperability between different systems and facilitates better communication by providing a common vocabulary and understanding. The primary advantage of using ontology models is their ability to represent complex information in an organized and accessible way, which improves the accuracy and efficiency of automated reasoning and decision-making processes [181]. They enable robust querying, analysis, and integration of data from various sources, thereby streamlining the assembly and disassembly tasks. However, there are also disadvantages. Developing comprehensive ontology models requires significant expertise and effort to accurately capture the intricacies of the domain. The process can be time-consuming and resource-intensive [182], involving meticulous detailing and regular updates to maintain relevance and accuracy. Additionally, the complexity of ontologies may lead to challenges in scalability and performance, particularly when dealing with large and dynamic datasets [183]. Despite these challenges, ontology models provide a powerful tool for enhancing the clarity, precision, and efficiency of object-oriented assembly and disassembly operations, facilitating better automation and integration.

The application of ontology models (Algorithm 3) in assembly and disassembly operations offers a robust framework for managing diverse products and handling uncertainties. A disassembly task planning approach based on partial destructive rules and ontology was proposed in [184] to establish a comprehensive model for automating the disassembly planning of end-of-life automotive power batteries. This approach combined rule-based reasoning and case-based reasoning as mechanisms for reusing and reasoning disassembly knowledge. When applied to large-scale heterogeneous automotive power battery disassembly, this method was able to generate disassembly task schemes efficiently and quickly. In [185], a general ontology model with a rule-based reasoning approach was introduced to generate the best disassembly scheme and sequence automatically. The proposed approach yielded an optimal disassembly scheme, outperforming other heuristic algorithms by minimizing disassembly direction changes and accomplishing the shortest process time. The reasoning procedure was easily trackable and modifiable, making the method universal, with the potential to support the whole remanufacturing process. Since automation solutions were predominantly tailored to specific applications and were not well equipped to handle the diverse range of products in disassembly environments, an ontology-based disassembly system was introduced in [186], incorporating distributed multi-agent machine control, robotic automation, as well as vision-based path planning. The integration of ontologies with agents endowed the system with flexibility and the capability to execute hardware and software modifications autonomously for specific tasks. The ontology semantic coupling facilitated the understanding of captured images, enabling robust and automated execution of disassembly operations and their coordination. Despite the challenges of developing and maintaining comprehensive ontology models, their ability to streamline operations and improve resilience to uncertainties makes them an invaluable tool in modern manufacturing.

Algorithm 3: Ontology-based object-oriented assembly and disassembly

3.5.2. Domain-Specific Languages

Domain-specific languages [187] provide a tailored approach to handling the unique requirements and complexities of the assembly and disassembly operation process. Domain-specific languages are specialized programming languages designed to address the specific needs of a particular domain, offering syntax and constructs that are closely aligned with the concepts and operations of assembly and disassembly tasks [188]. The primary advantage of domain-specific languages is their ability to enhance productivity and precision by providing clear and concise ways to express domain-specific logic, reducing the potential for errors and improving the maintainability of the code. Domain-specific languages facilitate communication between domain experts and developers, ensuring that the implemented solutions accurately reflect the intended processes. However, there are also disadvantages. Developing and maintaining domain-specific languages can be resource-intensive, requiring significant expertise in both the domain and language design. The initial investment in creating a domain-specific language might be high, and it may take time for developers to become proficient in using it. Additionally, domain-specific languages can lead to fragmentation if different teams or projects adopt different specialized languages, potentially complicating integration and interoperability [189]. Despite these challenges, domain-specific languages offer a powerful means to streamline and optimize object-oriented assembly and disassembly operations by providing tools that are specifically designed to handle the domain’s unique challenges and requirements.

The utilization of domain-specific languages in assembly and disassembly operations presents a powerful approach to handling errors through reverse execution and reducing development efforts. As the complexity of robotic assembly tasks increased, the likelihood of error-free execution diminished. Therefore, a system was proposed in [190] for automatically managing specific types of errors rather than solely preventing them. Numerous operations could be reversed automatically, and errors were addressed through the automatic reversal to a safe point. In [191], a programming model enabling the reverse execution of robot assembly programs was introduced. The challenges associated with executing robot programs backwards were investigated, and the classification of reversibility characteristics was presented. It was demonstrated how reversing program execution direction temporarily could serve as an effective error recovery mechanism. The initial design of a reversible assembly sequence domain-specific language was presented in [192] for specifying assembly sequences. Certain types of errors were addressed with reverse execution during assembly operations, enabling the robot to back out of an error temporarily, after which the operation could be retried automatically. Reversibility was utilized to generate an assembly sequence from a provided disassembly sequence automatically. While the initial creation of domain-specific languages requires significant expertise, their long-term benefits in error handling and development efficiency make them a valuable asset in industrial automation.

3.5.3. Planning Domain Definition Language

PDDL [193] provides a structured and formal approach to defining and solving complex planning problems. PDDL allows for the precise specification of the tasks, goals, actions, and constraints involved in assembly and disassembly processes. This facilitates the use of automated planners to generate optimal sequences of actions, enhancing efficiency and reducing the likelihood of errors [194]. The primary advantage of PDDL is its ability to handle intricate planning scenarios systematically, enabling high levels of automation and optimization. It supports a clear representation of the problem space, which can be crucial for tasks requiring detailed and logical planning. However, there are also disadvantages. Developing PDDL models requires substantial expertise in both the domain and the intricacies of PDDL syntax and semantics. The complexity of creating accurate and comprehensive models can be high, particularly for large-scale or highly detailed tasks [195]. Additionally, the computational demands of solving PDDL-based planning problems can be significant, especially as the complexity of the task increases. This can lead to longer processing times and the need for robust computational resources. Despite these challenges, PDDL remains a powerful tool for enhancing the planning and execution of object-oriented assembly and disassembly operations, offering significant benefits in terms of precision and optimization.

The application of PDDL (Algorithm 4) in assembly and disassembly operations provides a sophisticated method for handling reconfiguration and reducing uncertainties. While advances have been made in automated planning systems, they remain unsuitable for runtime robotic executions in uncertain environments. In [196], a PDDL-based simulated execution and robotic planning framework was provided, offering integration of orchestration, adaptive deployments, and automated planning. Structured re-planning rules in response to execution failures, state changes, and unforeseen obstacles were included. The implementation of classical programming and automated planning tools within a robotic disassembly and assembly system was detailed in [197]. The proposed method allowed for generating multiple initial and goal snapshots pertinent to real-world scenarios, enabling the resolution of various problems without the need for reprogramming. The integration of PDDL with robot language facilitated intelligent behavior in various industrial systems utilizing robots, offering benefits such as reduced uncertainties, adaptability, and behavior prediction. In [198], a PDDL-based assembly sequence generation approach was presented for robot programming. This method reduced the complexity of generating assembly sequences for products. Different from previous methods, this technique generated assembly sequences that were more aligned with the concept of reconfigurable assembly systems for general-purpose robots, enabled better evaluation for robotic assembly, and more accurately reflected assembly sequences in real-world systems. While developing and implementing PDDL-based systems require expertise in planning algorithms and domain modeling, the benefits of improved adaptability, reduced operational uncertainties, and streamlined process management underscore the value of PDDL in modern industrial automation.

Algorithm 4: PDDL-based object-oriented assembly and disassembly

3.5.4. Tree-Based Methods

Tree-based methods (Figure 13) leverage structured approaches such as decision trees [199] and RRTs [200] to enhance decision-making and path planning. Decision trees provide a straightforward and interpretable way to model decisions based on various criteria, making it easier to map out the steps required for assembly or disassembly [201]. This clarity helps identify the most efficient sequences of actions and understand the impact of each decision. RRTs are particularly effective in navigating complex and high-dimensional spaces, making them ideal for planning paths and movements in intricate assembly tasks [202]. RRTs can efficiently explore large spaces to find feasible paths, even in dynamic and unpredictable environments [203]. However, these methods also have disadvantages. Decision trees can become overly complex and may overfit the data, leading to less generalizable models that perform poorly on new or unseen tasks [204]. The simplicity that makes decision trees easy to understand can also limit their capability to handle highly nuanced or interdependent decisions. RRTs, while effective for path planning, can be computationally demanding, requiring substantial processing power and time, especially in highly cluttered or extensive environments [205]. Additionally, the random nature of RRTs can sometimes result in suboptimal paths if not carefully managed and tuned [206]. Despite these challenges, tree-based methods offer valuable frameworks for improving the efficiency and adaptability of object-oriented assembly and disassembly operations.

The application of tree-based methods (Algorithm 5) in assembly and disassembly operations presents a powerful strategy for reducing operational time and enhancing efficiency. A task planner based on decision trees was presented in [207]. The planner utilized a hierarchical product representation to distribute tasks using decision trees. By coordinating tasks, the system effectively reduced total disassembly time when performing parallel operations. The authors in [208] addressed planning processes for simulating industrial tasks including disassembly, assembly, or maintenance. An interactive path-planning algorithm based on an RRT-Connect method was presented. A real-time planner was proposed, enabling both a human operator and a computer to search the workspace simultaneously, thereby significantly accelerating the process. In [209], a novel motion planning approach for virtual disassembly and assembly operations was introduced. The rough path was retracted using a random retraction approach and then connected through the BiRRT algorithm. By leveraging this random retraction method, the approach facilitated efficient collaboration between the user and the automatic planner, significantly reducing assembly task planning time. A path planning algorithm based on prior path reuse (PPR algorithm) was introduced in [210], achieving rapid assembly path planning by reutilizing previously planned paths. The essence of the PPR algorithm was a strategy for path reuse, accomplished by enhancing the RRT* algorithm. The total planning time for assembly path planning using the PPR algorithm was significantly reduced compared to the RRT* algorithm (Figure 14). Despite the need for advanced computational resources and algorithm tuning, the substantial time savings and improved process optimization make tree-based methods invaluable tools in modern industrial automation.

Algorithm 5: RRT algorithm [200]

The application of tree-based methods in assembly and disassembly operations significantly enhances system adaptability. Different prediction methods based on ensemble learning and decision trees were presented in [211] to indicate the steps for the next assembly. The predictor was designed as a sensor-based assistance system, aiming to offer support through adaptive instructions. The novelty lies in the decision tree-based assembly state prediction, with ensemble learning using decision tree components being particularly well suited for adaptive assembly systems. In [212], an integrated planner based on the RRT algorithm for assembly tasks involving possible re-grasping was presented. Due to changes in the task space at each assembly step, nodes or edges in the tree could collide during the assembly of subsequent parts, disconnecting the colliding node from the entire tree. To address this, a two-stage extended RRT approach was proposed and a lazy collision-checking method was employed, allowing the reuse of the same tree throughout the assembly process instead of creating a new tree at each stage. A general framework for simultaneous assembly or disassembly path planning and sequencing was described in [213]. The developed algorithmic solution extended RRT, an efficient and straightforward path-planning algorithm. This approach was straightforward to implement and avoided the need for complex geometric computations. In [214], a computational path planner in intricate environments was introduced for assembly path planning. An innovative adaptive RRT algorithm was developed to address path-planning challenges in complex and uncertain scenarios. The path planner utilizing the proposed algorithms computed assembly paths with a higher success rate and greater efficiency in these demanding environments. Despite requiring sophisticated computational resources and careful algorithmic design, the increased adaptability and improved operational flexibility underscore the value of tree-based methods in advancing modern industrial automation.

4. Discussion and Future Research Directions

4.1. Discussion

Fusion power holds the promise of providing a nearly limitless and clean source of energy by mimicking the processes that power the sun. Utilizing isotopes of hydrogen, such as deuterium and tritium, fusion reactions produce immense amounts of energy with minimal radioactive waste compared to traditional nuclear fission [215]. However, maintaining a fusion reactor is a formidable challenge. Regular maintenance is crucial to ensure the reactor’s structural integrity and efficiency, often requiring advanced robotics and remote handling technologies to perform repairs and inspections in hazardous environments. Decommissioning necessitates meticulous planning and execution to safely manage and dispose of residual radioactive materials, ensuring minimal environmental impact. The future of fusion power depends not only on overcoming these technical challenges but also on making the entire lifecycle, from maintenance to decommissioning, safe, sustainable, and economically viable.

Object-oriented assembly and disassembly operations (

Table 5. Reviewed publications grouped according to the aforementioned methods.

Methods	Advantages	Disadvantages	References
Hardware Design and Creation	Handling delicate or complex parts. Ensuring accurate placement. Reducing the risk of damage.	Time-consuming. Costly. Limited versatility. Complex design.	[44,45,46,47,48]
Traditional Algorithms	Reliable and predictable. Reducing operational costs. Simplifying complex information. Mostly robust and deterministic. Accurate when handling spatial data.	Limited flexibility and adaptability. Time-consuming. Difficult and complicated. May converge on local optima. May require significant time for tuning and experimentation.	[60,61,62,63,68,69,70,71,72,73,74,76,81,83,84,85,90,91,92,96,98,99,101,102,103,110,111,112,113]
Neural Networks	Capable of processing and learning from vast amounts of data. Reducing task complexity. Adaptability to new tasks without explicit programming. Capable of processing unstructured data, such as images and sensor data. Flexible and scalable.	Requiring substantial computational resources and large datasets. Time-consuming. Costly. Low traceability such as hard to interpret the decision-making process. Difficult in debugging and fine-tuning.	[122,123,124,128,129,130,131,137,138,139,140,141,142]
Human–robot Collaboration	Rapidly transferring expert knowledge to robots. Quick reprogramming and customization. Improving product quality. Reducing production time.	Requiring significant effort from human experts. Costly. Resource-intensive and complex. Requiring sophisticated control systems to ensure safe interaction. Requiring additional training for human operators.	[150,151,152,153,157,158,159,160,161,162,163,167,168,169,170,171,172,173]
Knowledge Representation	Enhancing clarity and precision. Interoperability in representing task-related knowledge. Straightforward and easy to understand. Applicable for dynamic and unpredictable environments.	Difficult to develop and maintain. Computationally demanding and complex. Time-consuming. Costly. Less generalizable models. May struggle with high-dimensional spaces. May result in suboptimal paths.	[184,185,186,190,191,192,196,197,198,207,208,209,210,211,212,213,214]

) could offer a transformative approach to tackling the challenges of future plant maintenance and decommissioning. By decomposing complex maintenance and decommissioning tasks into smaller, manageable, and modular components, each designed with specific attributes and functions, this approach would allow for greater precision and flexibility. Advanced robotics and automated systems would efficiently assemble and disassemble these modular objects, performing intricate operations within the harsh and hazardous environments of a fusion reactor. This could reduce human exposure to radiation and minimize the risk of errors. Furthermore, the object-oriented framework could promote the development of standardized components and procedures, which can be reused and adapted across different reactors, streamlining maintenance processes and making decommissioning more systematic and predictable. This methodology will not only enhance safety and efficiency but also contribute to cost savings and sustainability, paving the way for the widespread adoption of fusion energy technology.

Methods in object-oriented assembly and disassembly operations within the application of plant maintenance will prioritize speed, precision, and safety far more than those applied in less critical contexts. Given the immense cost and complexity of fusion reactors, maximizing operational time while ensuring safety is paramount. This necessitates rapid and accurate maintenance processes to minimize downtime, which justifies substantial investments in advanced technology, automation, and processing power. In contrast, object-oriented methods outside the fusion context (i.e. manufacturing) often focus more on cost-efficiency and flexibility, where the conditions are less extreme, and the urgency for quick turnarounds is desired but less critical. In fusion applications, the potential consequences of damaging the vacuum vessel or other critical components due to slow or imprecise maintenance are significantly higher. Any damage to critical components like the vacuum vessel would result in exorbitant costs that far exceed those associated with the advanced technology, processing power, or time investment used to facilitate the fusion. Therefore, the willingness to allocate more resources to develop sophisticated, high-speed, and reliable systems is justified by the need to protect the expensive infrastructure and ensure continuous, safe operation of the reactor. The unique demands and high stakes of fusion reactor maintenance drive the need for these resource-intensive methods.

4.2. Future Research Directions

Future research in the field of plant maintenance and decommissioning could focus on several key areas to address the unique challenges and high stakes involved. Suggestions for future research directions are provided in the following areas:

• Development of High-Speed Robotic Systems for Plant Maintenance. Research into advanced robotic systems capable of performing maintenance tasks with high precision and speed. These systems would need to operate in high-radiation environments, utilizing remote handling and automation to reduce human exposure and error.
• Advanced Materials for Fusion Reactor Components. Investigation into new materials that can withstand the extreme conditions inside a fusion reactor, including high temperatures and neutron bombardment. These materials should enhance the durability and lifespan of reactor components, reducing the frequency and complexity of maintenance tasks.
• Real-Time Monitoring and Diagnostic Tools. Development of sophisticated monitoring systems that provide real-time data on the condition of fusion reactor components. These tools would use sensors and artificial intelligence to predict when and where maintenance is needed, allowing for proactive instead of reactive repairs.
• Artificial Intelligence and Machine Learning for Predictive Maintenance. Research into artificial intelligence and machine learning algorithms that analyze operational data to anticipate possible breakdowns and enhance maintenance schedules. This would help in minimizing downtime and ensuring the continuous, safe operation of fusion reactors.
• Modular Component Design for Easier Maintenance. Designing fusion reactor components in a modular fashion to simplify assembly and disassembly. This approach would facilitate quicker and more efficient maintenance procedures, reducing operational downtime.
• Improved Decommissioning Strategies and Technologies. Researching new methods and technologies for the safe and efficient decommissioning of fusion reactors. This includes developing techniques for handling and disposing of radioactive materials and dismantling reactor components.