The premise of this project is to train an agent via reinforcement learning to track an object with on-board camera. Below is a demo showing this in-action.
The simulation is done with Mujoco and the agent was trained with PPO. The files for training the agent are located in the "TRACKING_train" folder. The remaining files are the code for deploying the agent onto the corresponding microcontrollers.
Done:
- Local coordinate tracking trained. Agent transfer to hardware successful
To-do:
- Prediction
- Projectile
- Self-play