Blind Hexapod Locomotion Over Rough Terrains using Parallel DRL

Overview

This repository provides the codes used to train a phantomx robot to walk on rough terrain using NVIDIA's Isaac Gym. It includes the scripts used for training as well as those used for evaluation of the obtained policies.

Maintainer: Tania Sofía Ferreyra
Affiliation: Laboratorio de Robótica y Sistemas Embebidos, Instituto de Ciencias de la Computación, UBA-CONICET Contact: tferreyra@icc.fcen.uba.ar

---

Installation

1. Create a new python virtual env with python 3.10.
2. Install Isaac Gym and Isaac Lab following the documentation's instructions. Make sure to include the installation of rsl_rl.
3. Clone this repository.
4. Install python extension hexapod_extension using pip.

Usage

1. Train:
   python hexapod-locomotion-isaac-sim/scripts/train.py --task=Isaac-Velocity-Rough-Phantom-X-v0
   • To run headless add --headless.
   • To save videos of the training add --video.
   • The trained policy is saved in hexapod-locomotion-isaac-sim/logs/<experiment_name>/<date_time>_<run_name>/model_<iteration>.pt. Where <experiment_name> and <run_name> are defined in the train config.
   • You can override values set int he configuration files by adding any of the following line arguments:
   • --task TASK_NAME: changes the task.
   • --resume: resumes training from a given checkpoint.
   • --experiment_name EXPERIMENT_NAME: name of the experiment to run or load.
   • --num_envs NUM_ENVS: number of environments to use.
   • --seed SEED: random seed.
   • --max_iterations MAX_ITERATIONS: maximum number of training iterations.
2. Play a trained policy:
   python hexapod-locomotion-isaac-sim/scripts/play.py --task=Isaac-Velocity-Rough-Phantom-X-Play-v0
   • By default, the loaded policy is the last model of the last run of the experiment folder.