P

praca_licencjacka

Multi-agent reinforcement learning (MARL) is a powerful tool for managing city traffic lights. However, current training environments are often tied to a single simulator, which makes them hard to modify and limits research flexibility. To solve this, we created a flexible, modular framework for traffic signal control that works with multiple simulators. By providing a shared interface for MOSS, CityFlow, and SUMO, our environment separates the RL agents from the underlying traffic simulation. This makes it easy to test different algorithms (IDQN, IPPO, MAPPO) and custom traffic light rules without changing the agent's code. Furthermore, by fixing the performance bottlenecks of the older RESCO TensorCell baseline, our system trains agents three times faster while keeping the learning stable and the simulation accurate.

Analysis of Rumour Spreading on Twitter (Bachelor thesis)

praca_licencjacka

praca_licencjacka