LightSim: A Lightweight Cell Transmission Model Simulator for Traffic Signal Control Research

Researchers Haoran Su and Hanxiao Deng have introduced LightSim, a new lightweight traffic simulator designed to accelerate AI research for traffic signal control. The tool addresses a critical bottleneck: traditional simulators like SUMO require hours for training and days for setup, severely slowing the iteration cycle needed for rigorous reinforcement learning experiments. LightSim tackles this by focusing on the core dynamics relevant to signal timing—queue formation and discharge—using the established Cell Transmission Model (CTM). The result is a pure Python, pip-installable package with standard Gymnasium and PettingZoo interfaces that dramatically speeds up development.

Technically, LightSim achieves a remarkable speed of over 20,000 simulation steps per second on a single CPU. In cross-simulator validation tests spanning single intersections, grid networks, arterial corridors, and six real-world city networks, LightSim preserved the performance rankings of both classical and AI-based traffic controllers generated in SUMO, while enabling training that is 3 to 7 times faster. The researchers have released it as an open-source benchmark complete with 19 built-in scenarios, 7 controllers, and full reinforcement learning pipelines. This move effectively lowers the barrier to entry for traffic control research, transforming a process that once took days of setup into one that can be started in minutes.