Self-Driving Car Project
An autonomous vehicle simulation using reinforcement learning in Unity
Using reinforcement learning to navigate a 2d road environments in Unity. I have been using curriculum training to advanced the agents ability to handle difficult tasks. I implemented a custom dynamic graph generator and A* path finding algorthim for nav data. This sensor data, paired with raycasts that simulate camera seeing the percentage of road within it's line of "sight" are enabling it to reach the moving destination target. Currently I have trained the agent successfully with a number of scenes. I want to stablize its performance, some adjustments to the agent, and how the scenes move mostly, and then my next steps are to code the training algorithms rather than use ML Agents.
- Implemented curriculum learning for basic navigation
- Trained using Unity ML-Agents with progressive difficulty
- Analyzed performance with TensorBoard visualizations
- Integrated real-time visualization tools