This project aims to solve the issue of traditional control approach.
Drawbacks of traditional control approach:
- Requires domain knowledge on robot’s dynamic model
- Demand high computation power
- Not generalized to new environments
Therefore, in this project, we are using Proximal Policy Optimization (PPO) in PyBullet and implement quadruped robot to learn how to walk or even run.
Training after different iterations
Training with different controller combination
