The Key Technologies of New Generation Urban Traffic Control System Review and Prospect: Case by China

Abstract

Due to the limitations of its technology and theory, the traditional traffic control system has been unable to adapt to the needs of new technology and traffic development and needs to be reformed and reconstructed. From the national scientific and technological research and development plan to the traffic control system development projects of relevant enterprises, the common problem is that the advanced signal control system plays an insufficient role in practical application. The existing signal control system excessively relies on the use of IT technology but ignores the basic theory of traffic control and the essential consideration of the traffic environment and optimal regulation of road traffic flow, which greatly limits the scientific and practical value of a traffic control system in China. This narrative review analyzes recent developments and emerging trends in urban traffic control technologies through literature synthesis spanning 2009–2025. With the rapid and large-scale development and application of new transportation technologies such as vehicle–infrastructure networking, vehicle–infrastructure collaboration, and automatic driving, the real-time interaction between the traffic controller and the controlled party has new support. Given these technological advances, there is an urgent need to address the limitations of existing traffic signal control systems. Transportation technology development must leverage rich traffic control interaction conditions and comprehensive data to create next-generation systems. These new traffic optimization control systems should demonstrate high refinement, precision, better responsiveness, and enhanced intelligence. This paper can play a key role and influence for China to lead the development of urban road traffic control systems in the future. The promotion and application of the new generation of urban road traffic signal optimization control systems will improve the efficiency of the road network to a greater extent, reduce operating costs, prevent and alleviate road traffic congestion, and reduce energy consumption and emissions. At the same time, it will also provide the entry point and technical support for the development of vehicle–infrastructure networking and coordination and the automatic driving industry.

1. Introduction

1.1. Research Background and Significance

Traffic engineering professionals often refer to traffic control as the cornerstone of the field, while simultaneously representing a worldwide challenge. Given the complex and highly stochastic factors affecting traffic control, traffic signal timing optimization applications cannot directly adopt mature artificial intelligence and reinforcement learning algorithms. Additionally, since traffic represents a highly systemic entity, signal optimization must extend beyond optimal traffic light timing to include comprehensive congestion mitigation strategies and macro-level planning, avoiding the limitations of standalone traffic control systems. With China’s numerous rapidly developing cities, constantly increasing road traffic demands, citizens’ requirements for high-quality road transportation services, and the rapid development and potential applications of advanced technologies, how to further construct and develop high-level “new generation urban traffic control systems” has become a research direction of critical concern for Chinese cities.

On the other hand, China has issued numerous supportive policies for the intelligent connected vehicle industry, providing a solid foundation for the intelligent development of the “controlled entity” in control systems. In June 2017, the National Standards Committee, Ministry of Industry and Information Technology, and others released the “National Vehicle-Network Industry Standard System Construction Guide (Intelligent Connected Vehicles)”. In February 2020, 11 national ministries and commissions, including the Development and Reform Commission, Ministry of Industry and Information Technology, Ministry of Science and Technology, Ministry of Public Security, Ministry of Finance, and Ministry of Transport, jointly issued the “Intelligent Vehicle Innovation and Development Strategy”, emphasizing the coordinated development of intelligence and connectivity. In September 2022, the Shanghai Municipal Government Office issued the “Shanghai Implementation Plan for Accelerating Intelligent Connected Vehicle Innovation and Development” (hereinafter referred to as the “Plan”), which specified that by 2025, Shanghai would establish a nationally leading intelligent connected vehicle innovation and development system. The industry scale would strive to reach CNY 500 billion, with vehicles equipped with combined driving assistance functions (L2 level) and conditional autonomous driving functions (L3 level) accounting for over 70% of new vehicle production, while vehicles with high-level autonomous driving functions (L4 level and above) would achieve commercial applications in limited areas and specific scenarios. In January 2024, five ministries including the Ministry of Industry and Information Technology, Ministry of Public Security, Ministry of Natural Resources, Ministry of Housing and Urban-Rural Development, and Ministry of Transport jointly issued the “Notice on Conducting Intelligent Connected Vehicle ‘Vehicle-infrastructure-Cloud Integration’ Application Pilot Work”, requiring adherence to the principles of “government guidance, market-driven, coordinated planning, and sequential construction” to establish a batch of urban-level pilot projects with identical architecture, unified standards, interconnected services, and reliable security, promoting intelligent roadside infrastructure and cloud control platform construction, enhancing vehicle terminal equipment rates, conducting intelligent connected vehicle “vehicle-infrastructure-cloud integration” system architecture design and multi-scenario applications, forming unified vehicle–infrastructure coordination technical standards and testing evaluation systems, improving road traffic safety assurance capabilities, promoting large-scale demonstration applications and new business model exploration, and vigorously promoting the industrialization of intelligent connected vehicles.

However, vehicles must operate on roads according to traffic rules, and even the smartest intelligent connected vehicle cannot perceive all variables in the overall traffic environment relying solely on its own capabilities; cooperative coordination yields superior system-wide performance compared to isolated vehicle optimization. Traffic control is one of the most important technical means for regulating traffic flow, improving congestion, enhancing safety, and even reducing energy consumption and emissions. Recent empirical studies have provided compelling evidence demonstrating the substantial benefits of advanced traffic control systems in urban environments. Wu et al. [1] conducted a comprehensive analysis of China’s 100 most congested cities, revealing that big data empowered adaptive traffic signals achieved remarkable performance improvements: peak-hour trip times were reduced by 11% and off-peak times by 8%, resulting in an estimated annual CO₂ reduction of 31.73 million tons. Despite requiring an implementation investment of USD 1.48 billion annually, the societal benefits—encompassing CO₂ reduction, time savings, and fuel efficiency enhancements—were quantified at USD 31.82 billion, demonstrating a remarkable benefit–cost ratio of approximately 21:1. Furthermore, Chen and Yuan [2] developed an optimization framework integrating macroscopic traffic models with emission estimation models for Xi’an city, which successfully reduced both vehicle travel times and emissions of four major pollutant categories through environment-friendly signal control strategies. These findings substantiate the transformative potential of intelligent traffic control systems in simultaneously addressing traffic efficiency, environmental sustainability, and economic viability challenges in urban transportation networks.

Its progress and development have always kept pace with information and computer technology, as well as systems science. Consequently, traffic control theory and technology have evolved from offline to real-time online adaptive control systems. Due to the limitations of previous cross-sectional traffic detection technologies (loops, video, etc.), traffic control still cannot intelligently prevent or address traffic congestion proactively. At the micro-level, it remains difficult to control traffic flow with high precision and optimality, such as providing dynamic feedback on the effectiveness of traffic coordination control, making it even more challenging to ensure the operational effectiveness of the entire road network. Road intersections frequently experience queue spillover, underutilization of green signal time, and frequent traffic safety hazards. Empirical evidence from arterial traffic data substantiates these operational challenges. Wu et al. [3] demonstrated through high-resolution event-based data analysis that poor signal coordination and inappropriate signal timing significantly constrain intersection capacity, while the Queue-Over-Detector (QOD) phenomenon leads to misidentification of traffic states, masking actual congestion conditions. Liu et al. [4] further validated that traditional queue estimation methods fail under congested conditions when detectors are persistently occupied, highlighting the inadequacy of current signal control approaches in managing oversaturated intersections.

Given the importance of traffic control systems for improving and safeguarding urban traffic, academia and industry have almost continuously conducted research and development to enhance traffic control system functionality and performance. Through comprehensive review research on new generation urban traffic optimization control systems and key optimization controls, this paper aims to support the development of new generation traffic optimization control technologies and application products. Recent advances in artificial intelligence have catalyzed transformative developments in traffic signal control, with multi-agent deep reinforcement learning emerging as a paradigmatic shift from traditional optimization approaches. Rosca et al. [5] demonstrated next-generation AI architectures integrating edge computing with cloud-based analytics, achieving real-time traffic optimization through multi-parametric frameworks that consider route quality indicators encompassing safety, efficiency, and user preferences. Almusawi et al. [6] provided comprehensive empirical evidence on autonomous vehicle integration effects, revealing that different driving behaviors significantly impact traffic metrics across varying penetration rates. Furthermore, Zhou et al. [7] and Song et al. [8] advanced multi-agent reinforcement learning methodologies through incentive communication mechanisms and counterfactual actor–critic frameworks, respectively, addressing scalability challenges in coordinated intersection control while managing complex spatiotemporal traffic dynamics. These AI-driven approaches represent fundamental departures from conventional signal timing optimization, enabling adaptive, learning-based control strategies that continuously evolve with traffic patterns. Particularly in response to the rapid development of mobile internet, Internet of Things, vehicle networks, big data, and artificial intelligence, as well as new traffic control demands (service-oriented control, proactive control, priority control for special traffic such as public transportation and emergency vehicles, heterogeneous traffic flow coordination control, corridor traffic control, adverse weather control, energy-saving and emission-reduction traffic control, complex road network traffic control, etc.), traffic control systems are undergoing revolutionary changes, producing leading effects in the new round of global scientific and technological and industrial development competition.

1.2. Review Methodology

1.2.1. Literature Search Strategy

This narrative review employed a multi-database literature search to identify relevant studies on new generation urban traffic control systems. The primary databases searched included Web of Science Core Collection, IEEE Xplore Digital Library, ScienceDirect, Engineering Index (EI Compendex), and CNKI (China National Knowledge Infrastructure) to ensure comprehensive coverage of both international and Chinese literature. Additional sources included conference proceedings from major transportation conferences (TRB Annual Meeting, IEEE Intelligent Transportation Systems Conference) and technical reports from research institutions.

1.2.2. Search Terms and Scope

The search strategy utilized combinations of key terms including the following: “traffic signal control”, “intelligent transportation systems”, “connected automated vehicles”, “vehicle-infrastructure cooperation”, “traffic optimization”, “urban traffic control”, “heterogeneous traffic flow”, and “smart traffic systems”. Chinese equivalent terms were used for CNKI searches. The temporal scope focused on publications from 2009 to 2025 to capture the evolution from traditional traffic control to intelligent connected systems, while foundational earlier works were included when historically significant.

1.2.3. Inclusion and Exclusion Criteria

Inclusion criteria: (1) Peer-reviewed journal articles and conference papers; (2) technical reports from authoritative institutions; (3) studies focusing on urban road traffic control systems; (4) research addressing connected/automated vehicle integration; (5) publications in English or Chinese languages.

Exclusion criteria: (1) Studies limited to highway or rural traffic control without urban applicability; (2) pure theoretical works without practical implementation considerations; (3) research focusing solely on vehicle design rather than traffic control systems; (4) duplicate publications or preliminary conference versions of journal articles.

1.2.4. Literature Selection and Synthesis Process

The literature selection involved initial screening based on titles and abstracts, followed by full-text review for relevant studies. Given the narrative nature of this review, the synthesis process emphasized thematic organization around key technological developments, implementation challenges, and prospects rather than quantitative meta-analysis. Particular attention was given to research addressing Chinese urban traffic characteristics and implementation contexts to better understand the applicability of various control strategies in complex urban environments with mixed traffic flows.

1.3. Paper Structure

In addition to the introduction, this paper consists of seven sections: Section 2 addresses the origins and development of traffic control systems; Section 3 examines urban road intersection traffic temporal and spatial collaborative optimization methods under traditional conditions; Section 4 discusses the transformation of traffic control mechanisms in new heterogeneous traffic flow environments featuring intelligent connected vehicles and automated vehicles; Section 5 explores new heterogeneous traffic control mechanisms for urban roads in intelligent connected environments from three dimensions: isolated intersections, arterials, and networks; Section 6 presents empirical research on new heterogeneous traffic control mechanisms; Section 7 outlines the development directions for new generation urban traffic control systems; and Section 8 provides a summary and outlook.

2. Origins and Development of Traffic Control Systems

Although technologies and measures for controlling traffic flow based on traffic signals date back over a century, traffic coordination control involving multiple traffic signals originated in the 1950s–1960s with the offline multi-period control system proposed by the UK’s Transport Research Laboratory, which embedded optimized signal timing based on survey data into signal controllers. However, such control systems struggled to adapt to dynamic traffic conditions and diverse dynamic demands, and regenerating control schemes required extensive traffic surveys each time, proving time-consuming, labor-intensive, and costly.

Traffic signals at each road intersection regulate the right-of-way (phases), duration (green time length), and cycle time for each approach and movement direction, controlling pedestrian, bicycle, and motor vehicle traffic flows. Furthermore, by coordinating the green onset time differences (offsets, green wave bands) between upstream and downstream signals at two or multiple intersections, coordinated control of multiple intersections is achieved. To further implement system control for an entire city, traffic engineers commonly divide the road network into several zones (sub-area division) for inter-zone coordinated control. Therefore, achieving integrated optimization and dynamic optimization control for an entire city represents an extremely complex systems engineering challenge, requiring not only dynamic and precise understanding of the entire road network’s traffic conditions and trends but also comprehensive traffic facilities and orderly traffic flow as a foundation. While electronic, information, and computer technologies are necessary, the more critical and core element should be specialized traffic control technology based on traffic engineering expertise.

With the rapid development of computer and electronic technologies, the 1970s witnessed the gradual development of Australia’s SCATS (Sydney Coordinated Adaptive Traffic System; a dynamic plan selection control system based on detector-collected data) and the UK’s SCOOT system (Split Cycle Offset Optimization Technique; a dynamic prediction optimization control system based on detector-collected data). Subsequently emerging systems from Italy, the United States, Germany, Japan, and other countries share fundamentally similar control principles.

In the mid-1980s, China established the “Urban Traffic Real-time Adaptive Control System” project as a major national “Seventh Five-Year Plan” scientific and technological research initiative. This joint research project, undertaken by Tongji University (system design and core models and algorithms), the 28th Research Institute of the Ministry of Electronics Industry (system platform and signal controllers), the Traffic Management Research Institute of the Ministry of Public Security (project management), and the Nanjing Traffic Police Detachment (project implementation), was implemented in Nanjing. This represented the first systematic independent development of a traffic control system by Chinese researchers, laying the foundation for China’s subsequent traffic control technology research and development. The project achieved recognition as a major scientific and technological breakthrough during the “Seventh Five-Year Plan” period [9].

Although the “Seventh Five-Year Plan” scientific and technological research achieved significant results in traffic control systems, due to the need for improved commercialization, traffic control systems in China’s major cities still predominantly relied on imported systems: Beijing imported SCOOT systems; Shanghai, Tianjin, and others imported SCATS; Shenzhen imported Japan’s Kyosan system; Wuhan imported a Spanish system; and Jinan and Xi’an imported American systems. Some Chinese traffic control system manufacturers also established “signal control systems” in second-, third-, and fourth-tier cities, which were generally semi-manual and semi-automated systems adapted to traffic police dynamically adjusting signal timing, with relatively low functionality and performance, considerably lagging behind international standards [10].

3. Urban Road Intersection Traffic Spatiotemporal Collaborative Optimization Methods

Urban road intersections are critical nodes within urban road networks, with their operational efficiency directly affecting the performance of the entire urban transportation system. As traffic demands continue to grow and vehicle technologies rapidly develop, traditional single-dimension control methods can no longer meet the complex requirements of modern urban traffic. Spatiotemporal collaborative optimization, as an intersection optimization approach that comprehensively considers temporal resources (signal timing) and spatial resources (lane allocation, vehicle trajectories), is gradually becoming an important pathway for addressing this issue.

Traditional fixed-time signal timing optimization methods primarily rely on traffic volume data collected by infrastructure detectors (such as inductive loops), which typically involve high maintenance costs and low coverage. With the development of probe vehicle technology, vehicle trajectory data provides richer traffic state information, opening new possibilities for signal optimization. Research by Ma et al. [11] and Wang et al. [12] demonstrate that optimization methods based on vehicle trajectories can achieve effects even under low penetration rate conditions. The hierarchical multi-objective optimization framework developed by Ma et al. [11] compensates for the limitations of low penetration rate probe vehicles through the “Same-ratio Principles”, while Wang et al. [12] proposed a probabilistic time–space diagram method, establishing connections between stochastic point-queue models and vehicle trajectories. These methods have performed excellently in practical applications, with implementation results in Birmingham, Michigan showing reductions in intersection delay and stop frequency by up to 20% and 30%, respectively.

Building upon these trajectory-based optimization successes, the rapid development of connected and automated driving technology has provided an even broader scope for intersection spatiotemporal collaborative optimization. Compared to traditional methods, collaborative control based on vehicle-to-everything (V2X) environments can simultaneously optimize signal timing and vehicle trajectories, achieving more efficient traffic management. Feng et al. [13] modeled intersection control as a two-stage optimization problem: the first stage optimizes signal timing through dynamic programming to minimize vehicle delays, while the second stage applies optimal control theory to optimize vehicle trajectories to reduce fuel consumption and emissions. Simulation results indicate that this collaborative optimization method can reduce vehicle delays and CO₂ emissions by up to 24.0% and 13.8%, respectively. Liu et al. [14] went further by proposing a single-layer joint optimization method that simultaneously optimizes signal phase durations and vehicle platoon accelerations without requiring pre-specified terminal conditions, outperforming traditional methods in average travel delay, capacity, fuel consumption, and emissions. These studies demonstrate the enormous potential of spatiotemporal collaborative optimization in improving intersection efficiency and environmental sustainability.

In real traffic environments, connected and automated vehicles (CAVs) and human-driven vehicles (HVs) will coexist for an extended period, forming heterogeneous traffic flows. This mixed environment imposes higher requirements on traffic control, necessitating simultaneous consideration of the different characteristics of both vehicle types. To address this challenge, researchers have proposed various innovative methods. Ma et al. [15] introduced a traffic control model based on shared-phase-dedicated-lanes (SPDLs), allowing left-turns and CAVs to share dedicated lanes and pass through intersections together with HVs during shared phases. This approach achieved significant improvements in intersection efficiency through a three-level optimization model (signal optimization, phase configuration, and CAV trajectory planning).

Fully leveraging CAV advantages to guide heterogeneous traffic flow represents another research focus. Zheng et al. [16] innovatively proposed viewing CAVs as “catalysts” for vehicle platoons, promoting controllable platoon formation through various “catalytic” modes such as cooperative, accelerated, or direct lane-changing. Their two-stage optimization model optimizes vehicle delays and emissions, respectively, with numerical experiments indicating good results even at 20% CAV market penetration rates, reducing average vehicle delays and emissions by 54.32% and 19.1%, respectively. Similarly, research by Dai et al. [17] combined lane assignment with signal-vehicle coupled control, achieving the most significant benefits at moderate traffic demand levels with low sensitivity to CAV market adoption rates, indicating its applicability throughout the CAV adoption process.

Notably, compared to direct CAV control, routing control represents another method for improving heterogeneous traffic efficiency. The multi-objective routing control method developed by Moon et al. [18] based on deep Q-networks distributes traffic flow by controlling driving directions at intersections, proving particularly effective in dynamic traffic environments by enabling smooth traffic flow along optimal routes and reducing automated vehicles’ travel distances, times, and waiting times. Yang et al. [19] approached from a behavior prediction perspective, proposing a hierarchical prediction framework considering traffic signals and interactive agents, combining discrete intention prediction and continuous trajectory prediction to support automated driving decision-making in complex urban scenarios.

Beyond signal timing and vehicle trajectory optimization, lane allocation represents an important means for improving intersection throughput efficiency. Traditional design methods treat lane marking patterns as fixed inputs, limiting signal timing optimization space. Wong and Heydecker [20] broke this limitation by incorporating lane markings as binary control variables into optimization models, achieving synchronized optimization of lane marking patterns and signal timing. Yan et al. [21] and Yang et al. [22] further investigated phase swap sorting strategies, allowing different types of vehicle movements to be reorganized through pre-signals, enabling all lanes to discharge vehicles during green phases, significantly enhancing intersection capacity. These studies demonstrate that flexible allocation of lane functions and collaborative optimization with signal timing can more fully utilize limited road space resources.

Research on urban road intersection spatiotemporal collaborative optimization has achieved significant progress, conducting systematic exploration across multiple dimensions from vehicle trajectory data application, intelligent connected environment collaborative control, and heterogeneous traffic environment adaptation to flexible lane allocation. Research indicates that even under low penetration rate conditions, trajectory-based optimization methods can achieve significant effects; spatiotemporal collaborative control in intelligent connected environments simultaneously optimizes signal timing and vehicle trajectories, achieving dual reductions in delays and emissions; innovative strategies in heterogeneous traffic environments can fully leverage limited CAV resources to influence overall traffic flow; while flexible lane configurations and coordinated optimization with signal timing further improve the utilization efficiency of limited spatial resources. With the continuous development of vehicle networking and autonomous driving technologies, future research will increasingly emphasize collaborative control strategies and multi-objective balancing under heterogeneous traffic flow conditions.

Table 1. Comparison of urban road intersection spatiotemporal collaborative optimization methods.

Optimization Method Category	Main Characteristics	Applicable Scenarios	Performance Improvement	References
Signal Optimization-Based on Vehicle Trajectory Data	Uses vehicle trajectory data for signal optimization; compensates for low penetration rates through statistical methods; establishes probabilistic models linking traffic queues with vehicle movements	Low penetration rate probe vehicle environments; real urban intersections	Reduces delay by up to 20%, decreases number of stops by 30%; significant implementation results in Birmingham, Michigan	Ma et al. [11]; Wang et al. [12]
Spatiotemporal Collaborative Control in Connected Vehicle Environment	Two-stage optimization (signal timing first, then vehicle trajectories); single-layer joint optimization (simultaneously optimizing phase lengths and vehicle platoons); utilizing optimal control theory to reduce fuel consumption	High connected vehicle penetration rate environments; isolated intersections	Vehicle delay reduction up to 24.0%; CO2 emissions reduction of 13.8%; improved throughput; no need to prespecify terminal conditions	Feng et al. [13]; Liu et al. [14]
Collaborative Control Strategies in Heterogeneous Traffic Environment	Traffic control based on Shared-Phase-Dedicated-Lane (SPDL); using CAVs as “catalysts” to promote controllable platoon formation; combining lane assignment with signal-vehicle coupled control	Mixed environment of connected automated vehicles and human-driven vehicles; medium–low CAV penetration rate conditions (around 20%)	Average vehicle delay reduction of 54.32%; emissions reduction of 19.1%; most significant benefits under medium traffic demand; low sensitivity to CAV market adoption rate	Ma et al. [15]; Zheng et al. [16]; Dai et al. [17]
Routing Control and Behavior Prediction	Multi-objective routing control based on deep Q-networks; hierarchical prediction framework considering traffic signals and interactive agents; combining discrete intention prediction and continuous trajectory prediction	Dynamic traffic environments; complex urban scenarios; autonomous driving decision support	Reduces autonomous vehicles’ travel distance, time, and waiting time; improves prediction accuracy; enhances decision-making safety	Moon et al. [18]; Yang et al. [19]
Flexible Lane Assignment and Phase Swap Strategies	Incorporating lane markings as binary control variables in optimization models; phase swap sorting strategy allowing different movement types to be reorganized through pre-signals; enabling all lanes to be used for discharging vehicles during green phases	Saturated or near-saturated intersections; urban intersections requiring increased capacity; fixed infrastructure environments	Significantly increases intersection capacity; achieves synchronized optimization of lane marking patterns and signal timing; more efficiently utilizes limited road space resources	Wong and Heydecker [20]; Yan et al. [21]; Yang et al. [22]

summarizes the technical characteristics, applicable scenarios, and performance improvements of different spatiotemporal collaborative optimization methods, providing a systematic framework for understanding current research achievements and future development directions.

Despite the promising theoretical advances in spatiotemporal collaborative optimization, several critical implementation challenges persist. Penetration rate dependency represents a fundamental limitation, as most optimization methods require minimum CAV penetration rates (typically 20–30%) to achieve meaningful benefits, creating a “chicken-and-egg” problem during the transition period. Communication reliability poses another significant challenge, with V2X communication subject to latency, packet loss, and security vulnerabilities that can compromise control effectiveness. The computational complexity of real-time optimization, particularly for joint signal-trajectory planning, often exceeds current infrastructure capabilities, forcing practitioners to adopt simplified heuristics that may not achieve theoretical performance. Additionally, mixed traffic flow unpredictability remains problematic, as human driver behavior modeling limitations can lead to optimization failures when actual traffic patterns deviate from model assumptions. Infrastructure upgrade costs for implementing advanced detection and communication systems present substantial barriers for widespread deployment, particularly in developing urban areas with limited budgets.

In summary, research on urban road intersection traffic spatiotemporal collaborative optimization has made significant progress, exploring methods to improve intersection operational efficiency from different perspectives. With the continuous development of vehicle networking and autonomous driving technologies, future research will increasingly emphasize collaborative control strategies under heterogeneous traffic flow conditions, adaptive optimization methods based on big data and artificial intelligence, and low-carbon traffic control strategies oriented toward sustainable development. Particularly during the transition period of gradually increasing CAV penetration rates, how to fully leverage CAV advantages while accommodating the traffic efficiency of conventional vehicles will represent an important research topic. Spatiotemporal collaborative optimization will also extend from isolated intersections to arterial and network levels, constructing more efficient, safe, and environmentally friendly intelligent transportation systems through comprehensive consideration of signal timing, lane allocation, and vehicle trajectories.

4. Control Mechanism Transformation in New Heterogeneous Traffic Flow Environments on Urban Roads

In traditional traffic signal control systems, the controlled entities (vehicle drivers) can only passively adapt to traffic signals. Building upon the vehicle–infrastructure networking capabilities introduced in Section 1, intelligent connected vehicles transform the fundamental control paradigm. This transformation, as previously outlined, enables the formation of closed-loop feedback mechanisms incorporating all system elements. Under new heterogeneous traffic flow conditions, three traffic control modes exist: signal control, intelligent connected vehicle guidance and induction, and indirect control of conventional vehicles through intelligent connected vehicles. Integrating these three control modes, and specifically studying and clarifying control mechanism issues under new heterogeneous traffic flow conditions, requires constructing a collaborative mechanism that accommodates the three control modes under large-scale traffic data and new heterogeneous traffic flow conditions. This approach can enhance the global benefits of traffic control systems, improving existing traffic control efficiency while tapping into traffic control potential and strengthening system functionality and performance [23].

Compared to traditional traffic control methods, traffic control mechanisms under new heterogeneous traffic environments have the following characteristics: real-time collection of travel path information; automatic calibration of model parameters through vehicle–infrastructure cooperation and intelligent connected technologies; implementation of proactive control through prompting information, driving assistance, and other measures; enriched traffic control connotations due to the introduction of new heterogeneous traffic environments, such as dynamic path guidance and lane-changing control, enabling real-time interaction between drivers (demand) and traffic signals (supply).

Vehicle–infrastructure, vehicle–vehicle (intelligent and non-intelligent vehicles), and vehicle platoon coordination might achieve system-level collaboration through self-organization under highly intelligent coordination subject conditions. However, this remains extremely difficult for a considerable period. Therefore, a significant degree of organized control represents an important means and measure for achieving coordination between vehicles and infrastructure, between vehicles, and among vehicle platoons, ensuring the system develops toward a manageable target direction through scientifically effective control. Consequently, traffic collaborative control under new heterogeneous traffic environments stems, on one hand, from the significant application demands of new heterogeneous traffic and its increasingly practical technological environment, causing continuous changes in traditional traffic control environments and concepts and inspiring the development of new technologies and systems. On the other hand, new traffic control requirements, such as responding to emergencies, special traffic control requirements, and coordination between controlled entities and their influencing factors, drive the demand for next-generation traffic control technologies. From a traffic system functionality and objectives perspective, smoothness and efficiency represent the two primary goals; from a traffic system service object perspective, satisfying special traffic (intelligent connected vehicles) service demands will inevitably affect the entire system’s benefits. Therefore, research on next-generation traffic signal coordination control systems for new heterogeneous traffic on urban roads requires urgent resolution [24].

4.1. Evolution and Theoretical Foundations of Traffic Control Mechanisms

Traffic control systems have evolved from isolated fixed-time control to networked coordinated control and then to adaptive control. Zhao et al. systematically reviewed the four-generation development process of road traffic control systems: first-generation systems primarily employed fixed-time control, second-generation systems began introducing actuated control, third-generation systems implemented adaptive control through computer networking, while fourth-generation systems are characterized by multi-source data-driven approaches and connected vehicle participation [24]. Figure 1 illustrates this evolutionary progression, highlighting how each generation of traffic control has been enabled by specific technological breakthroughs, from mechanical timing devices in early systems to current 5G and edge computing technologies that enable vehicle–infrastructure cooperation.

The limited information volume of fixed-location traffic data collection methods constrains traditional adaptive control systems (such as SCOOT, SCATS, etc.), with unguaranteed precision and inability to obtain key information such as turning flows, link generation, and attraction volumes. Consequently, predicted traffic flow demands significantly differ from actual conditions, preventing traffic control from truly efficient real-time responses to traffic flow changes [25,26]. Chen et al. analyzed the application of Shanghai’s “Intelligent Traffic Lights” and the Sydney Coordinated Adaptive Traffic System (SCATS), noting that Shanghai’s “Intelligent Traffic Lights” offered higher real-time performance and stronger emergency response capabilities, yet still had significant room for improvement [27].

With the rapid development of artificial intelligence, Internet of Things, big data, and other technologies, traffic control systems are undergoing revolutionary changes. Li et al. proposed the concept of traffic cybernetics, systematically explaining the common principles and methods of traffic system control, arguing that information equals control, and proposing traffic cybernetics thinking that takes enhancing the system organization level as the fundamental goal, traffic entropy as the basic criterion, and information as the control means [28].

In terms of theoretical methods, Li et al. systematically analyzed the application of deep reinforcement learning in traffic signal control, discussing its advantages and challenges in complex traffic systems [23]. With the rapid development of vehicle networking technology and artificial intelligence algorithms, traditional model-based traffic control methods are gradually transitioning to data-driven intelligent control methods. The intelligent traffic signal control method based on deep reinforcement learning proposed by Wei et al. achieved real-time optimization of signal timing by integrating multiple traffic data sources, significantly improving intersection efficiency [29].

The application of edge computing and distributed control architectures provides new approaches for addressing the real-time performance and reliability challenges faced by traditional centralized control systems. Jin et al. developed a Fuzzy Intelligent Traffic Signal (FITS) control system that receives signal controller hardware messages through intermediate hardware devices and overrides traffic light indications during real-time operations, offering a convenient and economical method for improving existing traffic light infrastructure [30]. This distributed control architecture not only enhances system response speed but also strengthens system resilience against abnormal situations such as network failures.

4.2. New Heterogeneous Traffic Flow Characteristics and Model Construction

These fundamental transformations in control mechanisms are enabled by advances in data acquisition capabilities. Vehicle–infrastructure networking environments enable acquisition of traffic data with broader spatiotemporal dimensions and more refined and comprehensive characteristics [31]. An important feature of new heterogeneous traffic flow is the mixed operation of conventional vehicles and intelligent connected vehicles, with significant differences in traffic flow characteristics under different penetration rates. Zhang et al. analyzed the research status of vehicle group collaborative decision-making from the perspectives of vehicle group collaborative decision-making mechanisms, collaborative decision-making methods, and typical application scenarios in vehicle–infrastructure cooperative environments, with research results providing references for the management and control of new heterogeneous traffic systems in vehicle–infrastructure cooperative environments [32].

Regarding heterogeneous traffic flow detection and identification, Wang et al. reviewed data collection research in urban traffic control, analyzing the evolution process from fixed non-identifiable data, mobile detection data, to fixed identifiable data, indicating that practical application factors such as changes in data error rates due to detection environment differences and changes in traffic flow arrival patterns represent challenges facing current research [33].

Zhou et al. proposed a trajectory control theory for connected automated traffic in their research, providing a theoretical foundation for vehicle coordination control in heterogeneous traffic flow environments [34]. This trajectory-level traffic control method will play an increasingly important role in future heterogeneous traffic flow environments, providing theoretical support for precise control and optimization.

In terms of multi-objective optimization, Xu et al. proposed a cooperative method for vehicle speed control and traffic signal optimization, achieving comprehensive optimization of intersection throughput efficiency and energy conservation and emission reduction through speed guidance for connected vehicles and signal timing optimization [35]. This method fully considers the trade-offs between traffic efficiency and environmental protection, providing new approaches for multi-objective control in heterogeneous traffic flow environments.

4.3. New Traffic Control Methods in Intelligent Connected Environments

With the development of intelligent transportation systems, research on new generation traffic control system technologies based on multi-source heterogeneous data has gradually unfolded [32]. Yang et al. reviewed urban road traffic signal adaptive control methods, systematically analyzing the technical characteristics and application status of three categories of methods: model-based traffic control, intelligent computing-based traffic control, and data-driven traffic control [30].

Vehicle–infrastructure coordinated traffic control technology provides new approaches for addressing the limitations of traditional control methods. Yu et al. proposed an optimization method integrating traffic signals and vehicle trajectories, significantly improving intersection throughput efficiency through joint optimization of intersection signal timing and vehicle trajectories [36].

In terms of technical applications, Feng et al. proposed an adaptive signal control system based on vehicle–infrastructure coordination, capable of real-time estimation of intersection turning flows and queue lengths, providing precise data support for traffic signal optimization [37]. Wu et al. proposed an autonomous intersection management method based on ant colony systems, achieving coordinated vehicle passage through distributed decision-making [38].

Priority control technology for special traffic demands has been thoroughly researched. Zuo proposed an urban transit signal priority control method based on the Sydney Coordinated Adaptive Traffic System, designing control strategies accommodating bidirectional transit signal priority requests [39]. Li et al. proposed an actuator-based adaptive transit signal priority control system capable of dynamically adjusting priority strategies based on bus operational status and link traffic conditions [40].

With the development of intelligent connected technologies, traffic control methods are undergoing revolutionary changes. From a control mode perspective, traffic control will no longer be limited to traditional signal control but will combine various approaches such as vehicle trajectory guidance and overall traffic flow organization to form diversified control modes. From a control architecture perspective, traffic control system architectures are evolving from traditional centralized approaches toward distributed and edge-oriented directions, with edge computing technologies substantially enhancing system real-time performance and resilience, realizing multi-level control architectures with “cloud-edge-terminal” collaboration. From a control objective perspective, future traffic control will increasingly emphasize balancing multiple objectives such as efficiency, safety, environmental protection, and fairness. From a control method perspective, artificial intelligence-based traffic control methods will find wider application, with advanced algorithms such as deep reinforcement learning and multi-agent systems playing important roles in complex traffic environments.

Overall, traffic control mechanisms under new heterogeneous traffic environments are undergoing fundamental transformations, with traffic control systems gradually progressing from “inaccuracies in measurement, computation, evaluation, and control operations” toward precision, intelligence, and collaboration, providing important support for constructing efficient, safe, and green urban transportation systems.

The transformation of traffic control mechanisms in new heterogeneous traffic environments represents a paradigm shift from reactive to proactive control paradigms. As elucidated in this section, the evolution from fixed-time isolated control to network-wide adaptive systems has been accelerated by advances in sensing, communication, and computational technologies. However, the practical implementation of these transformative control mechanisms necessitates hierarchical considerations across different spatial scales of urban road networks. The mechanisms must be tailored to address specific operational characteristics and control objectives at various network levels while maintaining system-wide coherence. This spatiotemporal scalability requirement leads us to a more granular examination of control mechanisms in intelligent connected environments. The following section systematically investigates these mechanisms through a three-tiered approach—isolated intersections, arterial corridors, and network-level systems—to provide a comprehensive understanding of how emerging technologies can be effectively deployed to optimize traffic operations across the entire urban traffic system hierarchy. By stratifying the analysis across these three dimensions, we can better capture the nuanced requirements and performance benefits of new heterogeneous traffic control mechanisms while ensuring their interoperability within an integrated control framework.

4.4. AI-Driven Multi-Agent Architectures for Distributed Control

The evolution toward intelligent connected environments necessitates sophisticated multi-agent architectures that address fundamental challenges in centralized training with decentralized execution paradigms. Recent advances in AI-driven traffic control have demonstrated innovative solutions to overcome traditional multi-agent system limitations.

Yao et al. [41] exemplified the application of multi-agent reinforcement learning in interdependent network optimization, achieving 17.5% performance improvement through simultaneous network optimization (SNO) frameworks that account for functional interdependencies typically ignored in conventional approaches. This methodology addresses the modeling complexity inherent in heterogeneous traffic systems where decision coordination across multiple agents is critical. Du et al. [42] tackled the communication and coordination challenges through fog-cloud reinforcement learning architectures, implementing graph attention Q-networks (GAQ) with entropy-balanced routing to manage large-scale urban networks. Their centralized planning–decentralized execution framework achieved 14–54% improvements in traffic speed and congestion reduction, directly addressing the scalability issues identified in traditional multi-agent systems.

Furthermore, Niroumand et al. [43] advanced the paradigm through white phase mobile control mechanisms that integrate pedestrian movements with CAV trajectory optimization using receding horizon methodologies. This approach demonstrates how AI-driven coordination can manage complex multi-modal interactions while achieving up to 27% delay reductions through adaptive signal control strategies.

These developments illustrate the transition from conventional rule-based coordination to learning-based adaptive architectures that continuously evolve through real-time interaction with dynamic traffic environments, addressing the core limitations of traditional multi-agent traffic control systems.

4.5. Critical Challenges in Heterogeneous Traffic Control

The transformation toward intelligent connected traffic control faces several fundamental challenges that limit practical deployment. Scalability limitations emerge when extending single-intersection successes to network-level implementations, as coordination complexity grows exponentially with system size. Robustness concerns arise from the brittle nature of many AI-based control algorithms, which may fail catastrophically under edge cases or adversarial conditions not present in training data. Standardization gaps hinder interoperability between different manufacturers’ systems, creating vendor lock-in situations and limiting system flexibility. Cybersecurity vulnerabilities in connected systems introduce new attack vectors that traditional traffic control systems never faced, requiring entirely new security frameworks. Equity implications of optimization algorithms may inadvertently favor certain traffic flows or user groups, potentially exacerbating transportation disadvantages for vulnerable populations.

5. New Heterogeneous Traffic Control Mechanisms for Urban Roads in Intelligent Connected Environments

The rapid development of intelligent connected technologies is leading major transformations in the traffic control field. Compared to traditional traffic control, intelligent connected environments enable real-time information interaction capabilities between controllers and controlled entities, making refined, proactive, and collaborative traffic control possible. This section systematically organizes and deeply analyzes new heterogeneous traffic control mechanisms in intelligent connected environments from three different levels—isolated intersections, arterials, and networks—revealing their core technical characteristics and development trends.

5.1. Traffic Control Mechanisms for Isolated Intersection Scenarios

As the basic unit of urban road networks, isolated intersections form the foundation of intelligent connected traffic control research. Leveraging the heterogeneous traffic flow characteristics established in Section 4.2, intelligent connected environments enable expanded control objectives. The mixed-operation challenges and penetration rate considerations previously discussed now inform specific control mechanism design.

Safety-critical control mechanisms represent an important research direction for intelligent connected isolated intersection control. With the emergence of heterogeneous traffic conditions, traffic safety faces new challenges. Zhao et al. proposed a Safety-critical Traffic Control (STC) strategy enabling intelligent connected vehicles to simultaneously maintain safe distances with both preceding vehicles and following human-driven vehicles [44]. This research innovatively applies Control Barrier Functions (CBFs) to provide formalized safety guarantees of collision-free behavior for closed-loop systems, validating the method’s effectiveness in achieving provably safe traffic through extensive numerical simulations. This approach is particularly suitable for heterogeneous traffic environments with low intelligent connected vehicle penetration rates, minimizing deviations from nominal stabilizing controllers while ensuring safety.

Regarding adaptive signal optimization, traditional fixed-time signal control struggles to address dynamic traffic demands. To tackle this issue, Cai et al. developed an adaptive signal controller based on Approximate Dynamic Programming (ADP), achieving three operational objectives: dynamic green time allocation, automatic control parameter adjustment, and rapid signal plan revision [45]. This method alleviates computational burdens by using approximations of dynamic programming value functions and reinforcement learning to update approximations, achieving substantial vehicle delay reductions compared to optimized fixed-time plans in computer simulations. This near-real-time processing capability provides important support for dynamic signal control in intelligent connected environments.

Schedule-driven collaborative control represents a frontier direction for isolated intersection control. The cooperative schedule-driven intersection control method proposed by Hu et al. combines intelligent connected vehicles with intersection scheduling agents, optimizing platoon arrival times through wireless communication to adjust vehicle speeds, thereby reducing cumulative delays [46]. This method’s core lies in establishing closed-loop feedback mechanisms between intersection schedulers and vehicles, not only optimizing signal timing but also actively adjusting vehicle behavior to achieve overall system collaboration. This research demonstrates that by leveraging the communication and automation capabilities of intelligent connected vehicles, decentralized schedule-driven traffic control can improve traffic flow efficiency in complex urban road networks.

Flexible facilities and reservation control provide entirely new approaches for isolated intersection intelligent connected control. Wu et al. conducted a systematic review of intersection control in intelligent connected environments, focusing on analyzing the development status and limitations of two major categories of methods: intelligent connected vehicle trajectory planning and joint intersection control [47]. Building upon this foundation, Azadi et al. proposed the Combined Flexible Lane Assignment and Reservation-based Intersection Control (CFLARIC) concept, offering comprehensive lane assignment possibilities with appropriate reservation-based intersection control [48]. In simulation testing, this method demonstrated clear advantages in both efficiency (delays and stop frequency) and safety (reduced conflict situations) compared to traditional fixed-time control and basic reservation control. This innovative framework fully leverages the dynamic configuration capabilities of spatial resources in intelligent connected environments, providing new approaches for future intersection control.

Despite significant progress in intelligent connected isolated intersection control research, several key challenges remain: first, the adaptability of different control mechanisms to different intelligent connected vehicle penetration rates; second, balancing heterogeneous traffic safety and efficiency; third, computational efficiency and real-time requirements of control strategies. Future research needs to focus more on the progressive evolution of control mechanisms under different penetration rate conditions and more efficient algorithm designs to support real-time decision-making in complex traffic environments.

5.2. Traffic Control Mechanisms for Arterial Scenarios

Arterial traffic control represents the intermediate level connecting isolated intersection and network-level control, extending the control mechanisms established in Section 5.1 to coordinate multiple adjacent intersections. While isolated intersection control provides the foundation for intelligent connected traffic management, practical implementation requires expansion to corridor-level coordination. Arterial traffic control represents the intermediate level connecting isolated intersection and network-level control, with its core challenge lying in coordinating control strategies across multiple adjacent intersections to achieve traffic flow optimization over larger areas. Under intelligent connected environments, arterial traffic control mechanisms have achieved breakthrough developments in the following aspects.

Hierarchical control architectures provide effective solutions for complex arterial traffic systems. Xiao et al. proposed a two-layer longitudinal control method targeting heterogeneous traffic environments (intelligent connected vehicles and human-driven vehicles) on signalized arterials, effectively addressing traffic control complexity through hierarchical design with upper-layer travel time optimization and lower-layer intelligent connected vehicle movement optimization across urban arterial links [49]. This method, based on vehicle physical constraints and the relationship between estimated arrival times and traffic signal timing, considers four control scenarios and achieves travel time reductions of up to 29.33%. Lee et al. further validated the effectiveness of hierarchical control architectures through large-scale experiments in actual road environments, using 100 longitudinally controlled vehicles as Lagrangian traffic actuators in the CIRCLES project with a hierarchical control architecture named “MegaController” [50]. This architecture, comprising an upper-layer centralized “Speed Planner” optimization control algorithm and a lower-layer distributed vehicle-mounted control system, successfully suppressed traffic flow instabilities and reduced energy waste. These studies demonstrate that hierarchical control architectures can simultaneously satisfy dual requirements for global optimization and local execution, particularly suitable for complex traffic control in arterial scenarios.

Spatiotemporal collaborative optimization strategies represent key technologies for arterial control. The dynamic temporal and spatial speed control framework developed by Li et al., specifically designed for partially connected automated vehicles on signalized arterials, achieved traffic efficiency improvements through dynamic regulation of vehicle speeds in spatiotemporal dimensions [51]. This method first proposes a speed control optimization approach considering signal status and queuing conditions to minimize stop frequency for both the vehicle itself and following vehicles; second, introduces a secondary speed control method based on dynamic control areas to guide intelligent connected vehicles to target positions; and finally, designs corresponding dynamic variable parameter models to optimize control area operational parameters under different penetration rates, minimizing total fuel consumption for all vehicles. Simulation results indicate that at a 30% penetration rate, this strategy reduced total stop delays by 60.9% and saved 6.5% of total fuel consumption.

The decentralized approach for connected automated vehicle trajectory optimization along signalized arterials proposed by Wang et al. simultaneously considers both longitudinal and lateral dimensions [52]. This method employs a two-stage model: the first stage roughly estimates the minimal travel time required for a single connected automated vehicle traveling along the arterial, while the second stage optimizes connected automated vehicles’ longitudinal and lateral behaviors by minimizing delay and lane-changing costs. A rolling horizon approach dynamically implements this model, adapting to changing traffic conditions. Numerical experiments on actual road arterials demonstrate that this model can reduce average stop delays for both connected automated vehicles and human-driven vehicles, improving overall heterogeneous traffic efficiency.

Queue impact consideration represents an important aspect of arterial trajectory optimization. He et al. noted that many previous vehicle trajectory optimization studies for signalized arterials neglected intersection queue impacts, potentially leading to suboptimal or infeasible solutions [53]. They proposed a multi-stage optimal control formulation considering dual constraints of vehicle queuing and traffic signal status to obtain optimal vehicle trajectories on signalized arterials. To facilitate real-time updates of optimal speed trajectories, this research also proposed a constrained optimization model as an approximate approach, improving solution efficiency. Numerical examples demonstrated the effectiveness of the proposed optimal control model and the efficiency of the solution method. This optimization approach considering queue impacts provides more practical solutions for intelligent connected control in arterial scenarios.

The core value of intelligent connected traffic control in arterial scenarios lies in achieving coordination among multiple intersections to improve overall traffic operational efficiency. Compared to isolated intersection control, arterial control must address more complex spatiotemporal coordination relationships, considering factors including queue spillover, green wave band design, and multi-point coordination. Future research should further focus on the evolution of arterial coordination strategies under different penetration rates and how to achieve optimal control of arterial traffic under limited intelligent connected resource conditions.

5.3. Traffic Control Mechanisms for Network-Level Intersection Scenarios

Network-level traffic control represents the highest tier of traffic control systems, scaling beyond the arterial coordination methods described in Section 5.2 to achieve optimization throughout entire road networks. While arterial control extends isolated intersection techniques to corridor-level coordination, network control must integrate both single-intersection safety mechanisms (Section 5.1) and multi-intersection hierarchical frameworks (Section 5.2) into city-wide optimization systems. The intelligent connected environment provides rich data support and collaborative mechanisms for network-level control, achieving theoretical and technological breakthroughs primarily in the following aspects.

Multi-objective network optimization frameworks are key methods for addressing road network complexity. Fang et al. proposed a multi-objective traffic signal control method based on network-wide agent coordinated reinforcement learning (NACRL), aiming to achieve a multi-objective balance among traffic safety, efficiency, and network coordination [54]. This method designed a reward mechanism that simultaneously examines traffic safety and efficiency indicators, adopting a centralized training–decentralized execution framework to solve the data transmission limitation problems in field implementation. More importantly, the algorithm utilizes attention mechanisms to dynamically capture spatiotemporal dependencies across complex arterial networks, providing support for multi-agent coordinated control. Experiments on large-scale synthetic traffic grids and real-world arterial networks demonstrate that this algorithm simultaneously improves performance in terms of traffic safety, efficiency, and network coordination, while enhancing algorithm convergence and interpretability.

Zhu et al. developed a coordinated learning approach for network-level signal control, modeling each intersection as an intelligent agent that learns optimal timing decisions through interaction with the traffic environment. By using the JTA to obtain exact inference of the best joint actions for all coordinated intersections, VISSIM tests on a network containing 18 signalized intersections demonstrate that this algorithm outperforms independent learning (Q-learning), real-time adaptive learning, and fixed-time plans in terms of network-level average delay, number of stops, and vehicle emissions. This method effectively addresses the complexity and randomness problems of traffic signal control in large-scale road networks [55].

Multi-agent collaborative frameworks provide strong support for network-level control. Kolat et al. proposed a traffic signal control method based on multi-agent deep Q-learning algorithms, introducing innovative reward concepts in multi-agent environments, achieving 11% reduction in fuel consumption and 13% reduction in average travel time. This research demonstrates the potential of reinforcement learning in improving traffic signal controller coordination, providing support for building more sustainable and efficient transportation systems [56].

Hu et al. applied deep reinforcement learning methods (Double Deep Q-Network) to train local agents, with each local agent learning independently to accommodate regional traffic flows and dynamics. After completing the learning process, a global agent is created to integrate and unify the action policies selected by each local agent, achieving traffic signal coordination. This method is validated through urban mobility simulation, significantly improving average vehicle waiting time and queue length compared to PASSER-V and pre-timed signal setting strategies. This multi-agent learning strategy of “local first, then global” effectively balances local autonomy and global coordination [57].

Intersection heterogeneity is a key challenge faced by network-level control. Bie et al. point out that existing deep reinforcement learning models are typically developed in homogeneous road networks, ignoring the impact of intersection heterogeneity on action selection [58]. To address this issue, they proposed a value decomposition-based spatiotemporal graph attention multi-agent deep reinforcement learning model (MARL_SGAT), which quantifies action execution improvements in different road network environments by designing heterogeneous correlation indices utilizing road network structural parameters and formulating heterogeneous reward functions. The model utilizes spatiotemporal graph attention networks to process complex spatiotemporal features of traffic flows in heterogeneous road networks and introduces dual networks to convert the individual–global–maximum constraint of action selection into a value range constraint of the action advantage function. Simulation results show that the MARL_SGAT algorithm outperforms seven baseline algorithms, especially in heterogeneous road network environments, significantly reducing average vehicle delays and stops, and increasing travel speeds in the controlled road network.

Knowledge sharing and communication mechanisms are core to multi-agent collaborative control. The knowledge sharing deep deterministic policy gradient (KS-DDPG) method proposed by Li et al. achieves optimal control by enhancing cooperation between traffic signals [59]. This method introduces knowledge sharing communication protocol, allowing each agent to access the collective representation of the traffic environment collected by all agents. In two experiments using synthetic and real-world datasets, compared with state-of-the-art reinforcement learning-based and conventional transportation methods, KS-DDPG demonstrated significant efficiency in controlling large-scale transportation networks and coping with fluctuations in traffic flow. Additionally, the introduced communication mechanism has been proven to speed up model convergence without significantly increasing computational burden. This knowledge sharing mechanism provides a more efficient collaborative approach for network-level traffic control.

The decentralized framework based on region-aware cooperative strategy (RACS) and graph attention network (GAT) proposed by Wang et al. effectively addresses the problem that reinforcement learning algorithms solely based on value or policy are not suitable for large-scale multi-intersection scenarios [60]. This method assigns global control to each local reinforcement learning agent or intersection, adopting the advantage actor-critic (A2C) algorithm that combines policy function and value function, which has good convergence ability and is applicable to continuous state spaces. By introducing a region-aware cooperative strategy based on graph attention networks, the spatial information of surrounding agents can be incorporated, solving the partial observability challenges brought by decentralized methods. Experiments on synthetic traffic grids and the real-world traffic network of Monaco city confirm that the RACS method has shorter queue lengths and less waiting time than existing A2C and Q-learning algorithms and can reduce the total vehicle travel time.

The main challenges faced by network-level intelligent connected traffic control include computational complexity of large-scale systems, control consistency in heterogeneous road network environments, and the balance of multi-objective optimization. Li et al. conducted a systematic review of adaptive signal control and coordination for urban traffic in a connected vehicle environment, pointing out that although connected vehicle-based signal control shows significant improvements over existing conventional signal control systems, there are still many issues and drawbacks to overcome [61]. Future research should further explore more efficient agent communication mechanisms, control algorithms better adapted to heterogeneous road networks, and more practical network-level coordination strategies to support the construction of next-generation urban traffic signal control methods and systems.

5.4. Section Summary

While research demonstrates significant theoretical benefits across all three control scenarios (isolated, arterial, network), substantial gaps remain between laboratory results and real-world deployment. Technology maturity variations across different control mechanisms create uneven deployment landscapes, with some technologies requiring years of additional development before practical implementation. Economic viability remains questionable for many advanced control systems, particularly when considering total cost of ownership including maintenance, upgrades, and personnel training. Regulatory and institutional barriers often impede deployment of innovative control mechanisms, as transportation authorities require extensive validation and may resist departures from established practices. User acceptance challenges may limit effectiveness even when technical performance is satisfactory, particularly for control strategies that alter familiar traffic patterns or require behavioral adaptations from drivers.

In summary, the new heterogeneous traffic control mechanisms for urban roads in intelligent connected environments exhibit comprehensive characteristics of multiple levels, multiple methods, and multiple objectives. From the perspective of single intersections, methods such as safety control, adaptive signal optimization, schedule-driven collaborative control, and flexible facility reservation control each have their own characteristics and can effectively improve traffic efficiency and safety under different CAV penetration rates. From the arterial level, hierarchical control architectures, spatiotemporal collaborative optimization, and trajectory optimization considering queuing effects solve the complex problems of multi-intersection coordination. From the network level, innovative methods such as multi-objective network optimization, multi-agent collaborative frameworks, heterogeneous network control, and knowledge sharing and communication mechanisms effectively address the challenges of large-scale traffic control. These studies collectively indicate that with the continuous development of intelligent connected technologies, traffic control is evolving from traditional passive response to active prediction, from single-point optimization to network coordination, and from single-objective to multi-objective balance.

Table 2. Comparison of typical research on traffic control in different scenarios under intelligent connected environments.

Control Scenario	Method Category	Main Features/Innovations	Key Technologies	Performance Improvements	References
Isolated Intersection Scenario	Safety-Critical Control	Enables connected automated vehicles to maintain safe distances with both preceding and following vehicles; provides formal safety guarantees	Control Barrier Functions (CBFs); closed-loop system safety guarantees	Ensures collision-free behavior under various traffic conditions; minimizes deviation from nominal stabilizing controllers	Zhao et al. [44]
	Adaptive Signal Optimization	Achieves dynamic allocation of green time; automatic adjustment of control parameters; rapid revision of signal plans	Approximate Dynamic Programming (ADP); reinforcement learning	Substantially reduces vehicle delays compared to optimized fixed-time plans; significantly reduces computational burden	Cai et al. [45]
	Schedule-Driven Cooperative Control	Combines connected automated vehicles with intersection scheduling agents; adjusts vehicle speed through wireless communication	Closed-loop feedback mechanism between intersection scheduler and vehicles	Optimizes platoon arrival times; reduces cumulative delay	Hu et al. [46]
	Flexible Facility Reservation Control	Full spectrum of lane assignment possibilities; appropriate reservation-based intersection control	Combined Flexible Lane Assignment and Reservation-based Intersection Control (CFLARIC)	Reduces delays and stops compared to traditional control; improves safety	Azadi et al. [48]
Arterial Scenario	Hierarchical Control Architecture	Upper layer optimizes travel time; lower layer optimizes vehicle movement; considers four control scenarios	“MegaController” architecture; combination of centralized and distributed control	Travel time reduction up to 29.33%; successfully suppresses traffic flow instability; reduces energy waste	Xiao et al. [49]; Lee et al. [50]
	Spatiotemporal Collaborative Optimization	Dynamically regulates vehicle speed in spatial and temporal dimensions; considers both longitudinal and lateral dimensions	Dynamic temporal and spatial speed control framework; rolling horizon method	Reduces total stop delays by 60.9% at 30% penetration rate; saves 6.5% total fuel consumption	Li et al. [51]; Wang et al. [52]
	Trajectory Optimization Considering Queue Impact	Considers dual constraints of vehicle queuing and traffic signal status; constrained optimization model	Multi-stage optimal control; approximation solution methods	Obtains optimal vehicle trajectories on signalized arterials; improves solution efficiency	He et al. [53]
Network-Level Scenario	Multi-Objective Network Optimization	Achieves multi-objective balance of traffic safety, efficiency, and network coordination; resolves data transmission limitations	Network-wide Agent Coordinated Reinforcement Learning (NACRL); Junction Tree Algorithm (JTA)	Simultaneously improves traffic safety, efficiency, and network coordination performance; enhances algorithm convergence	Fang et al. [54]; Zhu et al. [55]
	Multi-Agent Cooperative Framework	Introduces innovative reward concepts; learning strategy from local to global	Multi-agent deep Q-learning; Double Deep Q-Network	Reduces fuel consumption by 11%; shortens average travel time by 13%; improves average vehicle waiting time	Kolat et al. [56]; Hu et al. [57]
	Heterogeneous Network Control	Considers impact of intersection heterogeneity on action selection; quantifies action execution improvements in heterogeneous road networks	Spatiotemporal Graph Attention Multi-Agent Reinforcement Learning (MARL_SGAT)	Significantly reduces average vehicle delays and stops in heterogeneous road networks; increases travel speeds	Bie et al. [58]
	Knowledge Sharing and Communication Mechanism	Knowledge sharing between agents; region-aware cooperative strategy; utilizes graph attention network	Knowledge Sharing Deep Deterministic Policy Gradient (KS-DDPG); Graph Attention Network (GAT)	Accelerates model convergence; enhances efficiency in controlling large-scale traffic networks; reduces queue length and waiting time	Li et al. [59]; Wang et al. [60]

systematically summarizes typical research methods, main innovations, key technologies, and performance improvements of traffic control in different scenarios under intelligent connected environments, comprehensively showcasing the research progress and technical systems in the field of new heterogeneous traffic control.

The future development of traffic control in intelligent connected environments should focus on the following aspects: first, exploring progressive control strategies under different intelligent connected vehicle penetration rates; second, developing more efficient and real-time algorithms to meet practical application requirements; third, researching coordinated control mechanisms in heterogeneous road network environments; fourth, constructing more comprehensive multi-objective optimization frameworks that balance efficiency, safety, energy consumption, and fairness among multiple demands. Through continuous innovation and in-depth research, traffic control in intelligent connected environments will provide powerful technical support for building more efficient, safer, and more environmentally friendly urban transportation systems.

6. Empirical Research on New Heterogeneous Traffic Control Mechanisms

Empirical research on control mechanisms in new heterogeneous traffic environments aims to verify the feasibility and effectiveness of various control strategies in actual or near-actual environments. With the development of connected automated driving technology, this type of research has gradually expanded from pure simulation environments to real road environments, while also actively exploring future control mechanisms oriented toward high penetration rates or fully automated driving environments. This section systematically analyzes existing empirical research findings to provide reference and guidance for the practical application of new heterogeneous traffic control mechanisms.

6.1. Testing and Validation of Control Mechanisms in Simulation Environments

Simulation environments provide a safe and controllable platform for the development and validation of control mechanisms, allowing researchers to systematically test the performance of different strategies under various scenarios. In recent years, simulation research for heterogeneous traffic environments has primarily focused on two directions: traffic flow stability control and intersection collaborative optimization.

6.1.1. Simulation Research on Traffic Flow Stability Control

Traffic flow instability (such as driving oscillations and stop-and-go phenomena) is one of the main causes of road congestion and energy waste. Multiple studies have shown that even at lower penetration rates, appropriately controlled CAVs can improve traffic flow stability.

Model Predictive Control (MPC)-based methods have performed excellently in stabilizing heterogeneous traffic flows. Liu et al. [62] and Li et al. [63] improved MPC applications in heterogeneous traffic environments from different perspectives. The Oscillation Mitigation-based MPC model (OM-MPC) proposed by Liu et al. effectively mitigates traffic oscillations even under shorter CAV distances (100 m) by innovatively estimating total equilibrium spacing. Li et al. expanded the perspective to bidirectional vehicle dynamic influences, developing a Robust Fuzzy MPC (RF-MPC) strategy that considers uncertainties in both preceding and following vehicles, using Takagi–Sugeno fuzzy models to capture human driving behavior, thus improving the overall stability and comfort of mixed platoons. These two studies collectively demonstrate that considering more comprehensive traffic environmental factors (such as oscillation characteristics and impacts from both preceding and following vehicles) can significantly enhance the effectiveness of MPC in heterogeneous traffic.

From a theoretical foundation perspective, Luo et al. systematically constructed a heterogeneous traffic stability analysis framework, proposing the Ring String Stability (RSS) concept, which allows string instability between predecessor–follower pairs provided they can be compensated by autonomous vehicles [64]. Simulation results validated this theoretical framework and quantified the upper limit of human-driven vehicles that a single AV can stabilize, providing a theoretical basis for optimizing AV deployment in heterogeneous traffic.

The common findings of these studies are as follows: (1) even at low penetration rates, rationally distributed CAVs can effectively stabilize traffic flow; (2) control effectiveness is closely related to CAV distribution strategies and control algorithms; (3) as external disturbances increase, the difficulty of maintaining stability also increases accordingly. These conclusions provide important references for traffic management strategies during the CAV transition period.

6.1.2. Simulation Research on Intersection Collaborative Optimization

As key nodes in urban traffic networks, intersections are particularly important for control optimization. In heterogeneous traffic environments, the collaborative optimization of signal control, vehicle trajectories, and spatial resources has become a research hotspot.

Multi-level control architectures are effective methods for handling the complexity of intersection collaborative optimization. Zou et al. [65] and Liu et al. [66] proposed collaborative control frameworks with different hierarchical structures. The two-level hierarchical framework developed by Zou et al. optimizes signal timing and vehicle arrival times at the upper level, and CAV trajectories at the lower level, achieving dual optimization of delay and energy consumption while maintaining computational efficiency. Liu et al. incorporated spatial resources (lane configurations) into the collaborative framework, achieving three-dimensional collaborative optimization of signal timing, lane configuration, and vehicle trajectories. These two frameworks differ in structure but both demonstrate the advantages of hierarchical optimization in balancing computational complexity and control effectiveness.

Platoon-based control methods that use leading CAVs to influence following vehicles are another effective heterogeneous traffic control strategy. The “1 + n” mixed platoon concept proposed by Chen et al., which influences n following human-driven vehicles through one leading CAV, achieves resource utilization maximization [67]. The research conducted rigorous stability and controllability analysis on the linearized dynamics model of mixed platoons, proving the superiority of this method compared to single CAV trajectory optimization, especially under medium to low CAV penetration rates.

For low penetration rate environments, trajectory data-based adaptive control strategies also show practical value. Wang et al. developed two simple yet effective adaptive control strategies relying only on sample delay and stop count data, avoiding dependence on expensive physical detection infrastructure [68]. This method validated the feasibility of low-cost sensing and control schemes in transitional period heterogeneous traffic management.

Synthesizing these simulation research results, we can identify the following commonalities: (1) collaborative optimization methods for heterogeneous traffic significantly outperform traditional control methods under various conditions; (2) hierarchical control architectures and platoon-based methods are effective approaches to addressing the complexity of mixed environments; (3) even under resource-limited conditions, simplified adaptive strategies can bring significant improvements. These findings provide practical guidance for the design of intersection control systems in heterogeneous traffic environments.

6.2. Testing and Validation of Control Mechanisms in Real Road Environments

While the simulation studies discussed above provide essential theoretical validation and practical guidance for intersection control system design, translating these findings to real-world implementation requires comprehensive field testing. Although simulation research provides important theoretical foundations, testing in real road environments is crucial for validating the actual effects of control strategies. In recent years, multiple research teams have tested heterogeneous traffic control mechanisms in real or semi-real environments, providing valuable data for translating theory into practice.

6.2.1. Testing of Vehicle-Level Collaborative Control

At the vehicle level, Cooperative Adaptive Cruise Control (CACC) is one of the most focused control mechanisms. Mu et al. utilized field data from the Netherlands CACC to compare and evaluate two CACC algorithms specifically designed for heterogeneous traffic: Linear Feedback Control (Linear-CACCu) and Adaptive Model Predictive Control (A-MPC-CACCu) [69]. Unlike traditional research focusing on highway scenarios, this study concentrated on urban arterial environments, with results showing the following: (1) both CACCu algorithms significantly outperformed conventional ACC and human driving in terms of safety, comfort, and energy efficiency; (2) traffic signal phase changes have a significant impact on optimal algorithm selection, suggesting dynamic switching of control modes based on expected signal phases.

Also based on CACC technology, Mu et al. extended their research to cooperative heterogeneous traffic platooning [70]. Their proposed framework utilizes communication between leading CAVs and intersections to optimize trajectories, with benefits transmitted through intermediate human-driven vehicles to following CAVs, forming a chain of cooperative effects. Following CAVs further optimize their behavior by receiving information from leading vehicles. This “lead-middle-follow” structure fully utilizes limited communication resources, achieving system-wide optimization even when only some vehicles possess communication capabilities.

These studies demonstrate that in real traffic environments, even when only some vehicles are equipped with connected automated driving capabilities, carefully designed collaborative control strategies can significantly enhance overall traffic performance. In particular, strategies utilizing leading CAVs to influence following vehicles show strong practicality and effectiveness in transitional heterogeneous traffic environments.

6.2.2. Validation of System-Level Collaborative Control

From a system-level perspective, vehicle–infrastructure collaboration is one of the core technologies for heterogeneous traffic control. The large-scale field test conducted by Calvert et al. in the Netherlands provided important empirical data. This test involved seven CACC vehicles traversing a provincial road corridor with five intelligent intersections, exchanging cooperative awareness messages and signal phase information through vehicle–infrastructure communication [71]. The study not only validated the technical feasibility of CACC platooning and vehicle–infrastructure communication but also predicted system benefits under different penetration rates through calibrated simulation models. The results showed that full CACC penetration can reduce average travel time by 5%, further improving to 11% when combined with vehicle–infrastructure communication technology, while at the provincial network scale, it can reduce total user delay by 12%.

Research by Wang et al. [68] validated the effectiveness of trajectory data-based adaptive control strategies in the field from another perspective. Through actual deployment testing, their two simple adaptive strategies significantly reduced delay, oversaturation, and spillover ratios, proving the feasibility of low-cost, high-efficiency heterogeneous traffic control solutions under existing infrastructure conditions.

These system-level test results reveal important conclusions: (1) vehicle–infrastructure collaboration technology can indeed significantly improve traffic performance in practical environments; (2) benefits typically increase with CAV penetration rates, with limited gains at low penetration rates below 10%; (3) even under limited conditions, simple but targeted control strategies can bring substantial improvements; (4) factors such as communication delays and signal interference in real environments have significant impacts on control effectiveness, which need to be fully considered in actual deployment.

Combining vehicle-level and system-level empirical research, we can see that heterogeneous traffic control mechanisms have shown good application prospects in real environments. Despite still facing challenges such as low penetration rates and unstable communication, a certain degree of traffic improvement has been achieved by combining intelligent vehicle control with infrastructure collaboration, accumulating valuable experience for larger-scale deployment.

6.3. Exploration of Control Mechanisms for Future Traffic Environments

With the continuous maturation of connected automated driving technology, researchers are also actively exploring future control mechanisms oriented toward high penetration rates or fully automated driving environments. Although these studies are still distant from large-scale practical applications, they provide important theoretical and technical reserves for the long-term development of transportation systems.

6.3.1. Autonomous Intersection Management Without Signals

In high CAV penetration rate environments, autonomous intersection management (AIM) without signals is expected to replace traditional traffic signal control, achieving more efficient intersection operations. The Trajectory-based Traffic Management (TTM) model proposed by Lu et al. [72] and the Decentralized Coordination Learning method (DCL-AIM) developed by Wu et al. [73] represent AIM technologies with two different approaches. The TTM model directly optimizes multi-vehicle trajectories through mixed integer programming and can be extended to consider scheduling decisions and fairness constraints; DCL-AIM models intersection control as a multi-agent collaborative decision-making problem, effectively reducing computational complexity through decomposing vehicle state spaces and dynamically adapting coordination needs. Both methods significantly outperform traditional control methods in different test scenarios, validating the enormous potential of AIM in future high CAV environments.

To further enhance the practicality of AIM, Hult et al. [74] and Ge et al. [75] optimized control architectures and algorithm efficiency from different perspectives. The bi-level model predictive controller developed by Hult et al. separates coordination control (time slot allocation) from vehicle control (trajectory planning), achieving control efficiency while ensuring safety. Experimental results prove that this controller can ensure safe coordination even under larger positioning uncertainties. Ge et al. focused on real-time performance, with their graph-based three-stage optimization method (target velocity optimization, vehicle subgraph extraction, and velocity profile synchronization) reducing single-step computation time to 0.02 s while maintaining control effectiveness, demonstrating the feasibility of AIM in real-time applications.

These studies collectively indicate that as CAV penetration rates increase, autonomous intersection management without signals will become an effective means to enhance intersection efficiency. The challenges lie in how to balance computational complexity, real-time performance, and control effectiveness, as well as how to handle practical issues such as positioning errors and communication delays. Methods such as hierarchical control, multi-agent learning, and graph theory optimization provide possible pathways to address these challenges.

6.3.2. Regional- and Network-Level Collaborative Control

Beyond single intersection control, regional- and network-level collaborative control is an important development direction for future transportation systems. The research by Lu et al. not only considered trajectory optimization at single intersections but also extended the scope to overall traffic management within dedicated autonomous vehicle zones [72]. By defining different fairness constraints in the TTMSE model, the research achieved a balance between two different operational objectives: system optimum and user equilibrium, providing a theoretical framework for regional-level collaborative control.

The large-scale testing and simulation conducted by Calvert et al. extended the assessment of network-level benefits [71]. The research showed that when CACC technology and intelligent intersection technology are applied to provincial-level road networks, they can bring a 12% reduction in total delay, validating the cumulative benefits of vehicle–infrastructure collaboration in network-level applications. This cross-intersection, cross-regional collaboration is an important feature of future traffic control systems, integrating single-point optimization into overall network optimization.

Synthesizing these exploratory studies oriented toward future traffic environments, we can summarize key trends: (1) expansion from single-point control to regional and network control; (2) evolution from single-objective optimization to multi-objective balance; (3) transformation from centralized control to distributed, self-organizing control; (4) development from passively adapting to traffic demand to actively guiding traffic behavior. These trends indicate that traffic control systems will undergo fundamental changes with the popularization of CAV technology, forming more intelligent, efficient, and sustainable future transportation systems.

6.4. Summary

Table 3. Comparison of empirical research methods for new heterogeneous traffic control mechanisms.

Evaluation Dimension	Research Focus	Main Results	Key Indicators	Existing Problems	Development Direction
Simulation Environment Research	Traffic Flow Stability Control	Oscillation Mitigation-based Model Predictive Control (OM-MPC)	Effectively mitigates traffic oscillations even with shorter CAV distances (100 m)	Low CAV penetration rate environments; intermittent traffic wave conditions	Liu et al. [62]
		Robust Fuzzy MPC (RF-MPC) Strategy	Considers uncertainties in preceding and following vehicles; improves overall stability and comfort of mixed platoons	Highly uncertain heterogeneous traffic flows; requires consideration of bidirectional vehicle dynamics	Li et al. [63]
		Ring String Stability (RSS) Analysis Framework	Quantifies the upper bound of human-driven vehicles that can be stabilized by a single AV; theoretically proves control feasibility under low penetration rates	Theoretical research; optimizes AV deployment strategies in heterogeneous traffic	Luo et al. [64]
	Intersection Collaborative Optimization	Two-level Hierarchical Framework	Upper level optimizes signal timing and vehicle arrival times; lower level optimizes CAV trajectories; achieves dual optimization of delay and energy consumption while maintaining computational efficiency	Limited computational resource environments; needs to balance computational complexity and control effect	Zou et al. [65]
		Signal-Lane-Trajectory Three-dimensional Collaborative Optimization	Incorporates spatial resources (lane configuration) into collaborative framework; achieves three-dimensional collaborative optimization	Needs comprehensive optimization of intersection spatial and temporal resources; isolated intersections	Liu et al. [66]
		“1 + n” Mixed Platoon Control Concept	One leading CAV controls and influences n following human-driven vehicles; maximizes resource utilization	Medium–low CAV penetration rate conditions; needs to influence non-CAV behavior	Chen et al. [67]
		Simplified Adaptive Control Based on Trajectory Data	Relies solely on sample delay and stop count data; avoids dependence on expensive physical detection infrastructure	Low-cost deployment environments; networks with limited detection facilities	Wang et al. [68]
Real Environment Research	Vehicle-level Cooperative Control	Linear Feedback Control (Linear-CACCu) and Adaptive MPC (A-MPC-CACCu)	In urban arterial environments, both algorithms significantly outperform conventional ACC and human driving in terms of safety, comfort, and energy efficiency	Urban arterial environments; requires dynamic switching of control modes based on expected signal phases	Mu et al. [69]
		Cooperative Heterogeneous Traffic Platooning Framework	“Lead-middle-following” platoon structure; benefits transmitted through intermediate human-driven vehicles	Mixed environments with partial vehicle communication capabilities; limited communication resource conditions	Mu et al. [70]
	System-level Cooperative Control	Large-scale Vehicle-Infrastructure Cooperation Field Test	Seven CACC vehicles traversing a provincial road corridor with five intelligent intersections; exchanging cooperative awareness messages and signal phase information through V2I communication	Full CACC penetration can reduce average travel time by 5%; combined with V2I communication can increase to 11%; provincial-level network can reduce total delay by 12%	Calvert et al. [71]
		Field Deployment of Adaptive Control Strategies Based on Trajectory Data	Actual deployment testing of two simple adaptive strategies; feasibility verification under existing infrastructure conditions	Significantly reduces delay, oversaturation, and spillover ratios; suitable for low-cost deployment	Wang et al. [68]
Future Exploration Research	Autonomous Intersection Management without Signals	Trajectory-based Traffic Management (TTM) Model	Directly optimizes multiple vehicle trajectories through mixed integer programming; expandable to consider scheduling decisions and equity constraints	High CAV penetration rate environments; can address complex intersection passage demands	Lu et al. [72]
		Decentralized Coordination Learning Method (DCL-AIM)	Models intersection control as a multi-agent cooperative decision problem; decomposes vehicle state space; dynamically adapts coordination needs	Effectively reduces computational complexity; applicable to complex intersection environments	Wu et al. [73]
		Bi-level Model Predictive Control Architecture	Separates coordination control (time window allocation) and vehicle control (trajectory planning); ensures safety	Ensures safe coordination even with larger positioning uncertainties	Hult et al. [74]
		Graph-based Three-stage Optimization Method	Target velocity optimization, vehicle subgraph extraction, and velocity profile synchronization; reduces single-step computation time to 0.02 s	Real-time application environments; requires high computational efficiency	Ge et al. [75]
	Area-level and Network-level Coordination	TTMSE Model with Different Equity Constraints	Achieves balance between system optimal and user equilibrium operational objectives; provides theoretical framework for area-level cooperative control	Dedicated autonomous vehicle zones; needs to resolve efficiency and equity balance issues	Lu et al. [72]
		Large-scale Network-level Benefit Assessment	Application of CACC technology and intelligent intersection technology to provincial-level road networks; integrates single-point optimization into network-wide optimization	Can bring 12% total delay reduction; verifies cumulative benefits of vehicle–infrastructure cooperation in network-level applications	Calvert et al. [71]

summarizes the main methods and results of empirical research on new heterogeneous traffic control mechanisms. Through systematic analysis of simulation environments, real environments, and exploratory research oriented toward the future, we can draw the following conclusions:

(1)

During the transition period with relatively low penetration rates, utilizing limited CAV resources to stabilize traffic flow and optimize intersection operations has already demonstrated technical feasibility. Through carefully designed control strategies, even a small number of CAVs can positively impact overall traffic.

(2)

Multi-level, distributed control architectures perform excellently in balancing computational complexity and control effectiveness, serving as an effective approach to address the complexity of heterogeneous traffic environments. Decomposing control problems into different levels or modules can achieve system collaborative optimization while maintaining computational efficiency.

(3)

Vehicle–infrastructure collaboration is a key enabling technology for heterogeneous traffic control. Real environment tests show that through information interaction between vehicles and infrastructure, system performance can be significantly enhanced, especially in structured environments such as signalized corridors.

(4)

Autonomous intersection management without signals and regional-level collaborative control oriented toward future high penetration rate environments demonstrate the revolutionary transformation potential of traffic control systems. Although these technologies are still distant from large-scale applications, they provide important theoretical and technical reserves for long-term development.

The comprehensive comparison tables presented throughout this review (Table 1, Table 2 and Table 3) provide essential technical and performance insights but reveal a critical gap in practical deployment assessment. Future research should systematically evaluate deployment readiness through standardized maturity frameworks, infrastructure requirement specifications, and cost–benefit analyses that are currently absent from the literature. This evaluation framework should distinguish between research prototypes, pilot-ready technologies, and commercially viable solutions while documenting the specific infrastructure investments, communication requirements, and system integration costs necessary for implementation. A critical limitation identified in our review is the absence of standardized performance metrics across different studies, preventing meaningful quantitative synthesis and meta-analysis. Future research should establish unified evaluation frameworks with normalized metrics for delay reduction, throughput improvement, energy consumption, and safety enhancement to enable systematic performance comparison across different control mechanisms and implementation scenarios.

Significant validation gaps exist in several critical areas identified through our review. System-level cooperative control lacks comprehensive real-world validation beyond the preliminary Dutch provincial road testing, leaving questions about scalability and robustness unaddressed. Future-oriented control mechanisms, including autonomous intersection management without signals and area-level coordination, remain almost entirely theoretical with minimal empirical validation. The transition from current low penetration heterogeneous environments (10–30% CAV penetration rates shown effective in simulations) to the high penetration scenarios (>70%) envisioned in future research represents the most substantial validation gap, with no empirical studies bridging this critical implementation threshold. Additionally, while individual control mechanisms show promise, integrated system validation combining multiple control approaches across different scales remains largely unexplored in real-world settings.

Based on the validation gaps identified above and the current state of technological development, we propose a structured research agenda with prioritized timelines. Immediate priorities (2025–2027) should focus on optimizing control strategies for low to medium CAV penetration rates (10–30%) and enhancing algorithm robustness in complex environments, as these address the most pressing near-term deployment challenges where simulation validation is mature but field testing remains limited. Medium-term priorities (2027–2030) should emphasize vehicle–infrastructure-cloud integrated collaborative control research and cross-scale multi-objective optimization frameworks, building upon the foundational work established in the immediate phase. Long-term priorities (2030–2035) should concentrate on larger-scale diversified road testing and validation of high penetration scenarios, representing the transition toward fully integrated intelligent transportation systems.

7. Development Directions of New Generation Urban Traffic Control Systems

Internationally, traditional traffic control systems tend to be well-established, forming mature traffic control system products developed mainly by traffic engineering and traffic control experts, which can particularly handle the improvement of road traffic and the coordinated development of traffic control systems. Ma Wanjing et al. conducted a comprehensive review of the progress and frontiers of intersection control for intelligent connected heterogeneous traffic flows, noting that signal timing oriented toward aggregate traffic flow in a single time dimension and trajectory planning oriented toward disaggregate individual vehicles in spatiotemporal dimensions are two control means for intelligent connected mixed flows that have complex coupling relationships, making collaborative optimization modeling difficult, with high dimensions and challenging efficient solutions, representing urgent difficulties to be overcome [76].

As documented in Section 2’s historical analysis, China’s traffic control system’s development has faced significant challenges with commercialization and practical implementation. Building on this historical context, current development efforts must address the fundamental issues identified earlier, particularly the gap between theoretical research and practical application. Effective improvement of the following issues in China’s traffic control systems and their technological development can achieve breakthroughs in key technologies of traffic control systems:

(1)

Enhance awareness of the specialized expertise required, including traffic flow modeling, real-time optimization algorithms, and vehicle–infrastructure communication protocols, recognizing the inherent complexity of multi-objective traffic control systems. Facing complex and changing traffic environments, the professional requirements for traffic control systems continue to increase. Majstorović et al. conducted a comprehensive literature review of urban traffic signal control under heterogeneous traffic flow conditions, pointing out the challenges and opportunities for traffic state estimation and traffic signal control in the coexistence of connected vehicles (CVs), connected automated vehicles (CAVs), and human-driven vehicles (HVs) [77]. With increasingly complex traffic composition, traffic control systems must possess enhanced adaptability and intelligence, requiring practitioners to have more professional knowledge backgrounds and technical capabilities. However, academia excessively conducts theoretical research detached from reality, making it difficult to apply learning to practice; the industry mainly develops products with IT focus (hardware such as signal controllers and detectors, with over 160 signal controller manufacturers in China that have passed standardized testing, but few can effectively implement networked control) and severely lacks professionalism; users emphasize hardware over software (preferring to invest heavily in hardware purchases with little investment in core algorithms and software systems for control systems), and the massive investments in traffic control system construction lack scientific post-evaluation.

(2)

Improve new road traffic facilities and their basic traffic flow passage environments. Road infrastructure in intelligent connected environments not only needs to meet basic passage needs but also adapt to the characteristics of new heterogeneous traffic flows. The model-based deep reinforcement learning method proposed by Wang et al. integrates traffic inference into the traffic signal control process, effectively improving control performance in complex traffic environments [78]. Research indicates that new road traffic facilities should possess the ability to perceive, analyze, and predict traffic states, providing reliable data support and execution guarantees for intelligent traffic control systems. However, for a long time, China’s road traffic construction has emphasized hardware over software and construction over management, with excessive civil engineering orientation and separation of construction and management, leading to unreasonable basic right-of-way for traffic flows, chaotic heterogeneous traffic flows, and difficulty in effective control implementation, greatly limiting the functionality of traffic control systems with limited effectiveness.

(3)

Promote the integration of industry, academia, research, and application in traffic control. The integration of industry, academia, research, and application is a key driving force for technological innovation in traffic control systems. The real-time signal priority coordination optimization method for self-driving buses at arterial intersections studied by Li et al. integrates theoretical models from academia with application requirements from engineering practice, fully considering the mutual influence between self-driving buses and private vehicles [79]. This research model oriented toward actual application scenarios helps promote the engineering transformation and application promotion of advanced traffic control theories, accelerating the organic integration of industry, academia, research, and application. However, academia, industry, research departments, etc., often independently conduct traffic control system technology research or development, which is an important reason why China’s traffic control systems lack theoretical foundation, hardware products cannot fulfill their proper roles, and traffic control system technology has not achieved substantial breakthroughs.

(4)

Construct a sustainable urban traffic control system service system. A sustainable urban traffic control system service system needs to integrate environmental protection concepts into control strategies. The eco-driving controller based on intelligent connected vehicles developed by Wang et al. optimizes road capacity and fuel consumption by combining vehicle dynamics and wireless communication technologies [80]. This control method, which considers both traffic efficiency and environmental protection, provides a practical path for constructing low-carbon, environmentally friendly urban traffic control systems, contributing to the sustainable development of transportation systems. As a city experiences continuous increases and dynamic changes in traffic demand, traffic problem countermeasures and improvements are long-term work. To fully leverage the role of massive investments in urban traffic control system construction, a high-level service team must be maintained.

In summary, there is a need to enhance understanding of the high professionalism and complexity of traffic control systems, improve controlled road traffic facilities and their basic traffic flow passage environments, promote the integration of industry, academia, research and application in traffic control, construct a sustainable urban traffic control system service system, and clarify the role of traffic control systems in improving traffic.

8. Conclusions and Discussion

8.1. The Limitations of the Review

This narrative review, while providing a comprehensive overview of new generation urban traffic control systems, has several inherent limitations that should be acknowledged for transparent academic discourse.

Methodological limitations include the narrative synthesis approach, which, although appropriate for this rapidly evolving interdisciplinary field, lacks the systematic rigor of formal systematic reviews with predetermined search protocols and standardized data extraction procedures. Our literature selection process, while extensive, may have inadvertently overlooked relevant studies due to the absence of exhaustive database searches across all possible sources.

Bias considerations present another set of limitations. Language bias is evident as our review primarily encompasses English and Chinese publications, potentially excluding valuable insights from research conducted in other linguistic contexts. Geographic bias toward Chinese traffic control developments, while intentional given our focus, may limit the global applicability of certain findings. Temporal bias toward recent publications (2009–2025) reflects the rapid technological evolution in this field but may have underemphasized foundational earlier works that remain conceptually relevant.

Scope limitations include our focus on urban road networks, which may not fully capture developments in highway or rural traffic control systems. The emphasis on connected and automated vehicle technologies, while reflecting current trends, may have resulted in less coverage of conventional traffic control innovations that remain relevant in many global contexts.

Technical limitations arise from the interdisciplinary nature of this field, spanning transportation engineering, computer science, and artificial intelligence. The rapid pace of technological advancement means that some cutting-edge developments may not yet be reflected in the peer-reviewed literature, creating a potential gap between current practice and published research.

Despite these limitations, we believe this review provides valuable insights into the evolution and future directions of urban traffic control systems, particularly in the context of intelligent connected vehicle integration and China’s unique urban development challenges.

8.2. Conclusions of the Review

Road transportation is the most critical foundation for a city’s economy, society, production, and daily life, yet the problems of urban road traffic congestion and blockage are becoming increasingly serious in most rapidly developing Chinese cities. Meanwhile, traffic control systems are undergoing revolutionary changes and have become one of the important fields in the new round of global transportation technology competition, with new generation road transportation technologies represented by vehicle networking, vehicle–infrastructure cooperation, autonomous driving, and traffic big data developing vigorously. However, China’s road transportation construction emphasizes hardware over software and construction over management, with excessive civil engineering orientation and separation of construction and management, leading to unreasonable basic right-of-way for traffic flows, chaotic heterogeneous traffic flows, and difficulty in effective control implementation, greatly limiting the functionality of traffic control systems with limited effectiveness. Facing major demands and objectives for future development, based on foreseeable scientific technologies, it is necessary to undertake traffic control system construction with leapfrog development, lead the development of China’s traffic control systems, and cultivate new technology industries for traffic control systems. However, current Chinese traffic control systems have problems such as inaccurate measurement, calculation, and evaluation, and are merely semi-manual, semi-automated systems adapting to traffic police dynamically adjusting signal timing, with low functionality and performance, significantly lagging behind international standards. Therefore, there is an urgent need to rapidly construct next-generation urban traffic optimization control systems, help China’s vast cities through substantial industry–academia–research–application cooperation, build advanced, practical, high-level traffic control systems in the true sense, and cultivate new technology industries for traffic control systems.

According to the current research status and development dynamics, numerous scholars and research institutions have conducted considerable research on implementation technologies, traffic state analysis, and experimental methods related to vehicle–infrastructure networking and collaboration, and have made preliminary explorations of traffic control methods for single intersections and special needs under vehicle–infrastructure collaboration conditions. However, these efforts primarily focus on physical system construction, adaptability analysis, and preliminary theoretical research, without yet forming collaborative control theories and experimental methods for new heterogeneous traffic flows. The vehicle-mounted terminals of intelligent vehicles can respond to signal control schemes, surrounding vehicles, and road network states, leading to significant changes in the operational mechanisms of new heterogeneous traffic flows; simultaneously, under new heterogeneous traffic flow conditions, vehicle position, speed, and other information can be collected more richly and accurately, and intelligent vehicles can be directly guided, thus forming three control modes: traffic signal control, intelligent vehicle guidance, and indirect control of conventional vehicles through intelligent vehicles. Therefore, their collaborative control mechanisms and principles urgently need research. This study originates precisely from the increasing popularization and application of intelligent vehicles, the significant changes in urban road traffic flows, and the urgent demand for collaborative control, possessing significant directionality, frontier nature, and trend significance. Related research can not only contribute to the innovative development of theoretical frameworks for new heterogeneous traffic flows on urban roads but also provide key support for countermeasures to urban road new heterogeneous traffic flow control problems and the development of next-generation traffic control systems.

In conclusion, this study is oriented toward the new heterogeneous traffic flow environment composed of human-driven vehicles, intelligent connected vehicles, and even autonomous vehicles that will coexist for a long time. This heterogeneous coexistence paradigm is supported by recent traffic flow research. Guo et al. [81] developed mathematical frameworks demonstrating that mixed HDV-CAV traffic networks require fundamentally different control strategies compared to homogeneous flows, while Yu et al. [82] empirically validated through car-following experiments that heterogeneous traffic exhibits distinct behavioral patterns that significantly impact system stability and capacity utilization at varying penetration rates. Through analyzing their basic characteristics and collaborative mechanisms, it establishes collaborative control theories and methods under new heterogeneous traffic environments and realizes collaborative control of new heterogeneous traffic flows oriented toward typical scenarios and application cases. Research on this issue has significant implications for deepening understanding and recognizing future new heterogeneous traffic flows, enhancing new heterogeneous traffic flow management and control capabilities, and advancing the development of intelligent connected vehicle collaborative control technologies oriented toward traffic efficiency in new heterogeneous traffic flows.