GETTING_STARTED.md

Getting Started

This document provides brief installation and training instructions.

• For general information, please see README.md
• For tasks descriptions, please see TASKS.md

Notes:

• The code has been tested with PyTorch 1.10, CUDA 11.3 and cuDNN 8.2
• All experiments in the initial paper were performed using IsaacGym Preview 2
• The code should be compatible with IsaacGym Preview 3/4 (not tested extensively)

Installation instructions

Create a conda environment:

conda create -n mvp python=3.7
conda activate mvp

Install PyTorch:

conda install pytorch torchvision -c pytorch

For RL experiments, install IsaacGym:

cd /path/to/isaac-gym/python
pip install -e .

Clone this repo:

cd /path/to/code
git clone [email protected]:ir413/mvp.git

Install Python dependencies:

cd /path/to/code/mvp
pip install -r requirements.txt

Install this repo:

cd /path/to/code/mvp
pip install -e .

Reinforcement Learning with PPO

Train FrankaPick from states:

python tools/train_ppo.py task=FrankaPick

Train FrankaPick from pixels:

python tools/train_ppo.py task=FrankaPickPixels

Test a policy after N iterations:

python tools/train_ppo.py test=True headless=False logdir=/path/to/job resume=N

Imitation Learning with BC

Steps:

1. Record demonstrations (see bc/dataset.py for expected format)
2. Train a policy on recorded demonstrations (example commands below)

Train on real demos:

python tools/train_bc.py logdir=/path/to/job/dir

Train on sim demos:

python tools/train_bc.py --config-name sim.yaml logdir=/path/to/job/dir