Due to globalization and the growth of urban areas, there are now more cars on the roads than ever before. As public transportation and cars use the same urban road traffic infrastructure, traffic congestion is the main issue facing almost every city. Urban roads are notorious for their traffic jams, which mostly happen at intersections where conflicting traffic flows are safely managed by traffic signals. In general, urban congestions can be divided into recurrent and non-recurrent. Recurrent congestions are primarily brought on by the physical limitations of infrastructure, daily recurring periods of increased traffic demand, and infrastructure management. Non-recurrent congestions are primarily brought on by traffic accidents, special events (sporting events, concerts, vehicle breakdowns, roadworks, etc.), and traffic incidents [1]. Recurring congestions are simpler to identify, and suitable traffic control measures can be prepared in advance to alleviate them. For the latter, accurate traffic state estimation is essential because identifying congestion is the first step in finding a solution to it. Successful congestion or its build-up detection allows for implementing suitable congestion-relieving measures, such as signal program changes or vehicle rerouting.

Frequent traffic congestions have an impact on daily living and present a variety of difficulties. Reducing traffic congestion minimizes environmental pollution while simultaneously enhancing travel efficiency and safety. Researching causes of congestion, the authors of [2] found that the most statistically significant relationships occurred in the case of the number of business entities and the number of passenger cars implying that congestion is more frequent in areas with a higher number of business entities. Air pollution and fuel loss are side effects that become severe issues as traffic congestion extends the time a vehicle is on the road. According to [3], the combined cost of traffic congestion in France, Germany, the United Kingdom, and the United States is expected to increase from USD 239.5 billion in 2020 to USD 293.1 billion by 2030. Thus, finding new ways to reduce traffic congestion is important for improving everyday life in urban areas.

Due to the widespread use of numerous Traffic Signal Control (TSC) systems worldwide, TSC is now a key element of Intelligent Transportation Systems (ITS) traffic control services. Many new technologies can be adopted in TSC and especially ITS. Artificial neural networks, fuzzy systems, and evolutionary computation algorithms are the core components of computational intelligence, which provides flexibility, autonomy, and robustness to handle the non-linearity and randomness of traffic systems [4]. Accurate modeling and short-term traffic prediction are quite challenging due to traffic’s intricate characteristics, stochastic, and dynamic traffic processes [5]. Due to their randomness and non-linearity, real-world data have limited applicability in short-term traffic state estimation ^[5][6][5,6].

Being able to solve complex real-world problems, in recent years, multi-agent systems also gained significant importance in the field of traffic engineering [7]. As in any domain, solving traffic engineering problems also requires domain expertise. Specially since relying on multi-agent systems, problems can be divided into multiple smaller problems that require less domain expertise. Due to this advantage, it offers in resolving complex problems with uncertainties, multi-agent systems have drawn a lot of attention in the field of TSC. These systems provide a highly flexible and modular structure that incorporates domain expertise to achieve the optimal solution [7].

The emerging mixed traffic flows enable even more control strategies for TSC. Mixed traffic flows are the result of the coexistence of conventional Human-Driven Vehicles and Connected Vehicles (CVs). One has to note that a CV is considered any vehicle, e.g., Connected Autonomous Vehicles (CAVs), and HDVs, being able to communicate with the environment whether that is another vehicle, infrastructure, or pedestrians [8]. Having the possibility to operate as mobile sensors, the appearance of CVs and CAVs has opened up new areas of research regarding ITS. Equipped with Radio Detection and Ranging (RADAR), Light Detection and Ranging (LIDAR), cameras, and many other sensors, CVs have significant advantages compared to conventional fixed-mounted traffic sensors. While conventional sensors cover only specific measurement points, each CV is a mobile data source that can provide real-time spatiotemporal measurement data. Thus, instead of having the traffic information for certain road sections, the data from CVs or CAVs can provide insight into the traffic state along the road on a microscopic level. Moreover, CVs or CAVs have advantages over existing traffic sensor technology because they are not limited by the line of sight like cameras and, as mentioned, collect large amounts of data at the microscopic level, which is convenient for studying traffic. Such large amounts of data will be generated by future mixed traffic flows containing classic vehicles and CVs or CAVs. The share of the latter will rise, decreasing the need for classic traffic sensors (inductive loops, cameras, radars, etc.).

Although CVs can provide data on the microscopic level, those data must be preprocessed before they can be used as input for various TCS systems. Hence, having a lot of data requires data processing to be fast and efficient. Therefore, the question of how to process large quantities of data quickly and efficiently using the potential of CVs and CAVs as mobile sensors and actuators arise. CAV-based multi-agent based traffic control systems are a possible solution as a single agent can process a small piece of information acting on it, and more agents together can handle very complex processes, including a network of intersections [9]. Thus, another question is the applicability of the CVs or CAVs in the multi-agent system since both can also receive information about the traffic state ahead.

2. Impact of Mixed Traffic Flows on Traffic Signal Control

The introduction of CVs alters traffic flow dynamics, increasing the effective road capacity as the penetration rate of CVs rises. Driven by advanced sensing and computing capabilities, and by utilizing V2X communication CVs can percept the environment. They also share onboard data (speed, position, and heading) in real-time with the environment through V2X communication. Thanks to sensing technology and real-time communication, vehicles are becoming also a good basis for developing new ITS-related applications and services. Exchanging traffic information CVs will have timely information about the traffic state and can recommend the driver’s ideal vehicle speed or alternative route. Relying on computation capabilities, CAVs will be part of a multi-agent system, where each vehicle will make decisions based on available data, which means CVs will affect traffic flow characteristics, increasing throughput and reducing delays. CVs are the vehicles with more information and recommendations, while CAVs are the vehicles with more information and orders. It is important to emphasize drivers’ compliance in CVs as they receive recommendations from the TSC contrary to CAVs that receive the orders. To enhance the performance of CV applications at low penetration rates, the authors of ^[10][60] proposed a new method to estimate the speeds and positions of CVs and non-connected vehicles. The proposed method utilizes the information collected from CVs and the speeds and flows of conventional vehicles collected from loop detectors. Thus, the proposed method estimates the forward movement of conventional vehicles based on collected data from CVs and sensors. Obtained results showed that estimation error increases as the CVs penetration rate decreases. Furthermore, estimation error decreases as the traffic demand decreases. For CVs penetration rates above 20%, the CVs-based TSC strategy outperformed the commercial EPICS adaptive control in terms of minimizing travel time delay and the number of stops. EPICS is an adaptive control system for individual intersection optimization developed by PTV Group. It calculates signal program parameters every second based on real-time traffic conditions, and depending on current conditions system decides whether the phase needs to be adjusted ^[11][61]. Results presented in ^[12][62] also indicated that the penetration rate of CVs above 20% improves traffic operations compared to existing approaches. The authors developed algorithms for mixed flow traffic state estimation, where the mixed traffic flow is contained of CVs and conventional vehicles without communication capabilities (both types of vehicles are HDVs). The CVs and sensors provided information about traffic, and the proposed methodology adjusted signal timing accordingly. Testing in a simulated environment on five scenarios indicated that even at the 10% penetration rate of CVs, the number of completed trips increased by 3.2%. From the above mentioned facts, it can be observed that the penetration ratio of CVs affects the traffic flow dynamics, although the CVs only provide onboard data. Being able to operate independently of the driver suggests that the CAVs could be used as the actuators in mixed traffic flows executing orders obtained from the TSC. Results from ^[13][21] indicated that the increase in road capacity happens gradually before the penetration rate of CAVs reaches 30%. Road capacity growth rate is mainly determined by the CAV capability on the desired time gap when the CAV penetration rate is over 30%. The traffic system performance improves as the percentage of CVs and CAVs increases. Thus, because their current market penetration rate is low, improving TSC is still challenging for traffic engineers to open new research/application areas. The use of CVs and CAVs is expected to have advantages such as increased road capacity, traffic safety, and efficiency. As mentioned above, acting as mobile sensors, CVs will provide rich real-time information to the TSC system. Based on provided data TSC system have better insight into the traffic state, and gathered information is used for optimizations of signal programs. Therefore, the TSC system can also inform drivers about the current traffic state, recommending the driver’s optimal speed profile and/or route. Currently, it is up to the driver if he will follow the recommendations. CAVs are also mobile sensors but provide even more information to TSC systems since CAVs are usually equipped with more sensors than CVs. Moreover, CAVs can be controlled by the TSC system, where the TSC system informs CAVs about signal phase timing and sequence. With such information, a particular CAV can alter its speed and route to reduce travel time and emissions. With a sufficient penetration rate, CAVs can affect the speed of other vehicles in traffic flow. However, a heterogeneous traffic flow made up of conventional HDVs, CVs, and CAVs will exist for a while before the CAVs are fully deployed, which could cause uncertainty in the existing transportation system. How much and in which way will the existing transportation system be affected is currently unknown. Furthermore, it is necessary to assess the connection between CVs and CAVs penetration rates and potential improvements in road capacity. A tabular overview of referenced papers regarding the impact of mixed traffic flows on TSC is summarized in the Table 1.