- Part 1: lane keeping with RL agent, after trained, collect images (1-8 cameras) with associated wheel angle and acceleration
- Part 2: Conv Net in TensorFlow, connect to Unity environment with Python API
- Input: Images collected from RL agent and (our driving)
- Targets: (wheel angle, torque, ...)
- Stretch goal: Use lidar (or radar) and add several RL agents