CBLAB: SCALABLE TRAFFIC SIMULATION WITH EN-RICHED DATA SUPPORTING

Abstract

Traffic simulation provides interactive data for the optimization of traffic control policies. However, existing traffic simulators are limited by their lack of scalability and shortage in input data, which prevents them from generating interactive data from traffic simulation in the scenarios of real large-scale city road networks. In this paper, we present City Brain Lab, a toolkit for scalable traffic simulation. CBLab consists of three components: CBEngine, CBData, and CBScenario. CBEngine is a highly efficient simulator supporting large-scale traffic simulation. CBData includes a traffic dataset with road network data of 100 cities all around the world. We also develop a pipeline to conduct a one-click transformation from raw road networks to input data of our traffic simulation. Combining CBEngine and CBData allows researchers to run scalable traffic simulations in the road network of real large-scale cities. Based on that, CB-Scenario implements an interactive environment and several baseline methods for two scenarios of traffic control policies respectively, with which traffic control policies adaptable for large-scale urban traffic can be trained and tuned. To the best of our knowledge, CBLab is the first infrastructure supporting traffic policy optimization in large-scale urban scenarios. The code is available on GitHub: https://github.com/CityBrainLab/CityBrainLab.git.



. These policies depend on data generated by interaction with the traffic environment where they explore to make good decisions under different consequences. However, real-world urban traffic cannot provide enough interactive data to train these policies, because the exploration of the policy may have a toxic impact on the urban traffic e.g. provoke severe congestion. Traffic simulators are therefore born as alternatives to provide traffic environments for traffic control policies to interact with. These simulators Lopez et al. (2018); Zhang et al. (2019); Chen et al. (2020b) simulate the microscopic evolution of the urban traffic. For each time step, they describe the traffic state, obtain a traffic action from the decision of traffic control policies and make it happen in the simulation. traffic control policies can then learn from how the traffic evolves under certain actions and improve decision making. While existing traffic simulators help hatch various traffic control policies successfully, they still come with drawbacks. Current simulators, as they were designed primitively, support simulation in road networks smaller than one hundred intersections and cannot scale to city-level traffic, which involves thousands of intersections. Due to limits in efficiency and mechanisms, these simulators are either not able to conduct a city-level simulation in a feasible time or set to prevent masses of vehicles from coming in the traffic. Another concern lies in the shortage of input data for large-scale traffic simulation. Although the map data of main cities in the world is now completed roughly and being refined, there is an absence of infrastructure for convenient access to the map data and a pipeline to transform it into simulation inputs. Therefore, inputs for traffic simulation only come from manual work and are limited to a small set of road networks Wei et al. To overcome two aforementioned drawbacks, we propose City Brain Lab, a novel toolkit for scalable traffic simulation. CBLab consists of three components: a microscopic traffic simulator CBEngine, a data tool CBData, and a traffic control policy environment CBScenario. CBEngine is of high efficiency which benefits from well-designed parallelization. With ordinary computing hardware, CBEngine is capable of running the traffic simulation on the scale of 10,000 intersections and 100,000 vehicles with a real-simulation time ratio of 1:4. CBData includes an accessible dataset that contains raw road networks of 100 main cities all around the world. A pipeline is prepared to automatically transform the raw data into input data for traffic simulation. Combining CBEngine and CBData, users can easily start up traffic simulation on real city-level road networks. Based on the scalable traffic simulation, we further implement CBScenario as an environment for two common traffic control policies: traffic signal control and congestion pricing. Users can design, develop, and train traffic control policies in the framework of CBScenario. To the best of our knowledge, we are the first to provide infrastructure for large-scale traffic control policy optimization. Our contribution can be summarized as follows. • We develop a scalable traffic simulator CBEngine which supports city-level microscopic traffic simulation for the first time. • We develop a data tool CBData to provide input data for large-scale traffic simulation. • Based on CBEngine and CBData, we implement an interactive environment CBScenario for two common traffic control policies under a large-scale setting.

2. CBENGINE: CITY-SCALE TRAFFIC SIMULATION ENGINE

In this section, we introduce the design of CBEngine. We demonstrate the modeling of urban traffic and conduct extensive experiments to show the efficiency and scalability of CBEngine.

2.1. THE TRAFFIC MODEL OF CBENGINE

The objective of the simulator is to describe the interaction between the road network and the traffic flow (vehicles in the traffic). As shown in Figure 2 , the road network involves the interaction through one of its components: the traffic signal lights, which control the passing of vehicles at intersections. When the simulation starts, vehicles in the traffic flow set out from their origins, travel down the routes, and finally arrive at their destinations. Specifically, traffic signals and vehicles interact as follows. Each traffic signal may change the signal phase (controlling the traffic direction allowed to move) as time changes. Vehicles move on, accelerating or decelerating according to the current speed, the traffic signal the movement of nearby vehicles, and other circumstantial factors. Meanwhile, they may also change their routes accordingly. For each time step, traffic signals, and vehicles observe from their views and make decisions on their next action a t (driving and routing). The simulator will then conduct these actions. This forwards the system from state s t at time t to a new state s t+δt at time t + δt. One such step can be concluded by Eq 1. Due to the page limit, concrete modeling of the road network, traffic signals and traffic flows is discussed in Appendix C. s t + a t → s t+δt (1)



Figure 1: An overview of CBLab.

(2019c;b); Zheng et al. (2019); Xu et al. (2021) whose scales are often dozens of intersections (e.g. 4x3 or 4x4) -much less than real urban road networks.

