This repository implements the LOFT approach described in the IROS 2021 paper:

Learning Symbolic Operators for Task and Motion Planning
Tom Silver*, Rohan Chitnis*, Joshua Tenenbaum, Leslie Pack Kaelbling, Tomas Lozano-Perez
Link to paper: https://arxiv.org/abs/2103.00589

Instructions for running (tested on OS X and Ubuntu 18.04):

• Use Python 3.6 or higher.
• Download Python dependencies: pip install -r requirements.txt.
• Download the NDR package to a location on your path: https://github.com/tomsilver/ndr

Now, ./run.sh should work, and should finish in less than a second. You should see the printout In total, solved 30 / 30 near the end. The three different environments can be run by changing the ENV variable in run.sh. Data has been included already in the data/ folder, but if you would like to regenerate it, you can set COLLECT_DATA=1 in run.sh. All three environments should yield 100% test success rate on all seeds.

For questions or comments, please email tslvr@mit.edu and ronuchit@mit.edu.