This repository contains code based on the paper Learning to Act by Predicting the Future by Alexey Dosovitskiy and Vladlen Koltun, adapted for use with the MINOS simulator.
Make sure you have the MINOS library installed on your system, and then follow the guide in dfp to install prerequisites for this code.
See the example experiment configurations in experiments. You can train an agent from inside a specific configuration directory using:
python3 run_exp.py train --env_config objectgoal_suncg_sf --gpu
Then, testing can be done through:
python3 run_exp.py test --env_config objectgoal_suncg_sf --gpu --test_checkpoint checkpoints/<timestamp>
By replacing test
with show
you can visualize the test episodes. Use python3 run_exp.py --help
to get usage information for additional command line arguments. Refer to the MINOS instructions for more details on parameters and environment configurations.