Whole-body PID + QP for Stanford Woofer

Overview

The goal of this project is to simulate a controller that uses whole-body PID and QP. The wholy-body PID generates desired body accelerations. A QP solver then allocates foot forces to achieve the desired body accelerations. Then, multiplication by the jacobian transpose will transform the foot forces into joint torques.

Install

1. Activate a python3 conda environment

source activate [environment name]

2. Install OSQP

conda install -c oxfordcontrol osqp

3. Install mujoco-py: https://github.com/openai/mujoco-py

Run

python3 SimulateWoofer.py

Details

Implementation

• QP solver: OSQP
• Simulator: MuJoCo

Lessons from "High-slope terrain locomotion for torque-controlled quadruped robots"

• Minimize joint torques using W=JcS⊤WτSJc⊤, where W is the positive-definite used to regularize f in equation (6).
• Regularize foot forces to be normal to surface, however, regularizing joint torques was more practical