ReadMe.md

Waypoint-Based Reinforcement Learning for Robot Manipulation Tasks

This repository provides our implementation of Waypoint-Based RL. The videos for experiments can be found here

Installation

Clone the repository using

git clone https://github.com/VT-Collab/rl-waypoints.git

Implementation of rl-waypoints

Navigate to the rl-waypoints repository using

cd rl-waypoints
cd rlwp

To train the robot for manipulation tasks in the Robosuite simulation environment, run the following commands

python3 main.py train=True task=<task_name>

The complete set of arguments can be found in the \cfg folder. The tasks for the training can be from the following: {Lift, Stack, NutAssembly, PickPlace, Door}

Testing the trained models

We provide a trained model for each task for our approach. The trained models can be evaluated as follows

python3 main.py test=True task=<task_name> run_name=test render=True