Quadruped_Robot_with_Reinforcement_Learning

This project aims to solve the issue of traditional control approach.
Drawbacks of traditional control approach:

1. Requires domain knowledge on robot’s dynamic model
2. Demand high computation power
3. 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.

Pipeline

Reward Function

Result

Training after different iterations

Training with different controller combination