Compare different algorithm for quadruped

1. introduction

1.1 classic control

• plan: body trajectory optimization(TO)
• control: optimized whole body control with null space projection(OP-NSP-WBC)
• dynamic: pinocchio
• optimal problem solver: casadi

1.2 reinforce learning

• algorithm: ppo
  • input: robot states
  • output: foot velocity

1.3 visualization

• rviz
  • urdf: robot
  • trajectory: body, foot
  • point: body, foot, zmp

2. install

2.1 dependency

• install anaconda or miniconda, and new environment named py37:

- git clone [email protected]:wanghg1992/quad_comp.git --recurse-submodules
- cd quad_comp/
- conda env create -f environment.yml

2.2 install gym-quad-env

- cd quad_comp/
- pip3 install -e simulation_environment/

3. run

3.1 run classic control

roscore
./run_classic_control.sh

3.2 run rl train

roscore
./run_baseline_ppo.sh

3.3 run visualization

roscore
./run_visualization.sh

3.4 test simulation environment

roscore
./test_simulation.sh