Do you live in a metropolitan area? What do you hate the most about your life? Probably the commute!! What seems to make it worse? The traffic lights could be the cause.
Traffic Signal Control is probably the most important & influential aspect of Traffic Management.
The research paper by Majid Raeis and Alberto Leon-Garcia titled “A Deep Reinforcement Learning Approach for Fair Traffic Signal Control” analyses standard traffic control and proposes a better way for the management of traffic signals.
Overhead traffic lights, Belfast. Image credit: Albert Bridge via geograph.ie, CC BY-SA 2.0
Importance of this Research
Authors have documented the purpose of the traffic Control in urban areas very aptly:
In recent years, traffic control methods based on deep reinforcement learning (DRL) have gained attention due to their ability to exploit realtime traffic data, which is often poorly used by the traditional hand-crafted methods. While most recent DRL-based methods have focused on maximizing the throughput or minimizing the average travel time of the vehicles, the fairness of the traffic signal controllers has often been neglected. This is particularly important as neglecting fairness can lead to situations where some vehicles experience extreme waiting times, or where the throughput of a particular traffic flow is highly impacted by the fluctuations of another conflicting flow at the intersection. In order to address these issues, we introduce two notions of fairness: delay-based and throughput-based fairness, which correspond to the two issues mentioned above. Furthermore, we propose two DRL-based traffic signal control methods for implementing these fairness notions, that can achieve a high throughput as well
The Objective of the Research
Two notions of fairness in Traffic Control Systems are introduced in the article. Parameters and features to consider while designing a traffic control system are:
Delay Based Fairness: To reduce the maximum waiting time of any single vehicle
Throughput based Fairness: To reduce the total wait time for the vehicles
The two parameters mentioned above form the basis of the researchers proposing two DRL-based algorithms.
A trade-off hyper parameter is introduced to adjust the balance between delay-based fairness & throughput-based fairness.
Baseline methods for traffic control
The developed technique was compared against these methods:
Self-organizing Traffic Lights (SOTL): The current phase of light can change only if the below conditions are met
Minimum time elapsed from last phase change
The number of vehicles in a queue > predefined threshold
The speed at which the number of vehicles that are added in a queue crosses a certain number
Max Pressure: This approach tries to minimize the difference between total queue lengths of the incoming and outgoing approaches
A four-way intersection where the west/east road segments (major roads) are 250 meters long each and have 3 incoming and 3 outgoing lanes, while the north/south segments are 200 meters long with 2 incoming and 2 outgoing lanes. Furthermore, the WE traffic flows have a speed limit of 50 km/h, while the NS flows have a speed limit of 30 km/h
The experiment is run on an urban traffic simulator SUMO.
Proposed Models & Experiment Results
The above image shows the Cumulative Distribution functions of the waiting time for the WE & NS traffic flows separately. For the WE traffic flow, DFC2 results in the lowest maximum waiting time. In SOTL, a larger number of vehicles experience extreme waiting times. On the other hand, DFC2 and max-pressure have the best performance for the NS/SN traffic flow. Therefore, both traffic flows considered, DFC2 has the best performance among the compared methods
The above image shows how TFC, Max-pressure and SOTL perform in this scenario. The throughput of the WE traffic flow is highly affected by the variations of the NS traffic for Max-pressure and SOTL controllers. Results also clearly show that TFC can minimize that effect. Furthermore, image (d-f) shows that our controller has a much better performance in terms of the average waiting times (for both WE and NS flows) than the Max-pressure and SOTL methods, which result in large fluctuations of the waiting times for the NS traffic flow.
Throughput based traffic Control (TFC). Image credit: arXiv:2107.10146
Experiments by the researchers confirmed the superiority of the proposed mechanism over the existing baseline algorithms. Implementing the proposed solution for urban areas would eliminate the need to deal with constant traffic control problems and would ensure fairness in terms of individual waiting time for all traffic participants while maintaining high throughput.
Source: Majid Raeis and Alberto Leon-Garcia titled “A Deep Reinforcement Learning Approach for Fair Traffic Signal Control” https://arxiv.org/pdf/2107.10146.pdf
You can offer your link to a page which is relevant to the topic of this post.