brachiation's Introduction

Learning to Brachiate via Simplified Model Imitation

This repo is the codebase for the SIGGRAPH 2022 conference paper with the title above. Please find the paper and demo at our project website https://brachiation-rl.github.io/brachiation/.

Prerequisites

Linux or macOS
Python 3.8 or newer

Install Requirements

Download and install custom PyBullet build

git clone [email protected]:belinghy/bullet3.git
pip install ./bullet3

Install required packages

[email protected]:brachiation-rl/brachiation.git
cd brachiation
pip install -r requirements.txt

Quick Start

Test installation is complete by running passive forward simulation

python run_full.py --mode test

This repo contains pretrained models and examples of generated trajectories. Run and visualize the pretrained full model controller

python run_full.py --mode play --net data/best_full.pt

Run and visualize simplified model controller

python run_simplified.py --mode play --net data/best_simple.pt

Run and visualize full model planning. This mode is slow when simulating on CPU; if needed, reduce the number of parallel simulations on L61 in run_full.py.

python run_full.py --mode plan --net data/best_full.pt

Training

Training simplified and full model from scratch

# Step 1: Train simplified and dump trajectories in data/trajectories/
python run_simplified.py --mode train
python run_simplified.py --mode dump --net <saved-model-file>

# Step 2: Train full model (uses previously saved trajectories)
python run_full.py --mode train

Citation

If you use this code for your research, please cite our paper.

@inproceedings{brachiation2022,
  author = {Reda, Daniele and Ling, Hung Yu and van de Panne, Michiel},
  title = {Learning to Brachiate via Simplified Model Imitation},
  year = {2022},
  publisher = {Association for Computing Machinery},
  booktitle = {ACM SIGGRAPH 2022 Conference Proceedings},
  articleno = {24},
  numpages = {9},
  series = {SIGGRAPH '22}
}

Recommend Projects