Deep Reinforcement Learning in Complex Traffic Environments
An intelligent autonomous vehicle system designed to navigate complex, dynamic traffic environments. Unlike traditional rule-based systems, this project utilises Deep Reinforcement Learning to enable the agent to learn optimal driving policies through trial and error.
A custom-built, realistic 3D environment in Unity. It simulates:
The agent utilises PPO, a state-of-the-art policy gradient method, to master continuous control. It processes discrete inputs (sensors, speed, traffic state) to output continuous steering and acceleration actions.
A modular reward architecture guides the learning process. The agent balances multiple objectives: