ShiBot-Inu

Overview

ShiBot-Inu is a custom built quadruped robot designed for simulation and reinforcement learning. This repository includes the URDF description of ShiBot-Inu, Isaac Sim assets, and the control scripts.

Simulation Demos

Training is done on thousands of robot environments in parallel to speed up training of the PPO algorithm.

After 5000 training iterations, the ShiBot-Inu was able to learn a fast walking gait.

Training and Play

The following command is used to train the robot on 1000 environments without GUI.

python scripts/rsl_rl/train.py --task=Template-Shibot-Inu-Rl-Direct-v0 --num_envs=1000 --headless

The following command is used to test the learned policy of the robot with GUI.

python scripts/rsl_rl/play.py --task=Template-Shibot-Inu-Rl-Direct-v0 --num_envs=1000