Unity-Based Autonomous Driving Environment: A Platform for Validating Reinforcement Learning Agents

Our aim in this research is to develop from scratch a unique racing environment in Unity with the goal of implementing and validating various reinforcement learning algorithms. Once the environment is designed, research efforts will be focused on exploring solutions through different RL (Reinforcement Learning) algorithms by modifying rewards, behaviors, and algorithms. To facilitate the implementation of a broader range of algorithms, a Wrapper will be created to allow testing of algorithms using the Stable Baselines 3 library. Additionally, the research will explore the application of different reinforcement learning techniques, such as imitation learning or curriculum learning. After achieving promising results with training a single agent, the research will explore the possibility of training multiple agents simultaneously in the environment to observe how agents learn to interact with each other. Finally, an instructional application will be developed to consolidate the knowledge generated by the various algorithms, enabling users to visually observe the agents' learning progress. In this application, each algorithm will be represented by a car, allowing users to clearly see the performance of different algorithms in races and laps around the track, highlighting their strengths and weaknesses. In summary, the entire research aims to create an instructive and interactive application where the majority of reinforcement learning algorithms can be visually validated in an environment specifically created for the research.