mando1106 / peg-in-hole-with-rl

Peg-in-Hole with RL algorithm

This project provides diff models to complete the assembly tasks (peg-in-hole) in soft body and rigid body environmnets. The RL-PPO algorithm is mainly used in this project and the task for rigid body environment is completed (the policy includes hole seaching, shape matching, insertion). For the soft part, it uses soft body (soft tool) to complete the insertion task.

Details and demos: https://0707yiliu.github.io/peg-in-hole-with-RL/

Description

This repo contains the whole RL framework but without the trained model. Running code in ./src, RL-env in gym_envs. You can use the command in ./src/command.txt to quick start a trainning.

Quick Started

• Register the gym-env (go to the repo root)

pip install -e .

• quick trainning

python /your/repo/root/src/run.py  -e URPiHDense-v1 -t 7000000 -a PPO -v vision-touch -lr 0.0003 -hs 4mm -l -dsl -dr -nor -g 3

Dependencies

• gym >= 2.1
• Mujoco >= 2.3
• Python >= 3.7
• KDL-pkg
• ...

Version History

• 0.2
  • Update soft body version
  • See commit change or See release history
• 0.1
  • Initial Release