Skip to content

etaoxing/domain-shift-benchmark

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

domain-shift-benchmark

Quickstart

You'll probably want to first install KitchenShift, the environment we use in our paper, forked from adept_envs.

pip install -r requirements.txt
pip install -e .

To run behavioral cloning:

python tools/run_experiment.py exp_compare_bc -p experiments/domain_shift_benchmark/ -s 0 1 2 3 -w 0 1 2 3 -v 0

This starts 4 runs on each of 4 GPUs for variant 0 from exp_compare_bc.py, see the _VARIANTS_ variable in the experiment file for a list of the models. The random seeds used are specified by _SEEDS_. The run logs will be located in _EXP_DIR + /{MODEL}/run_seed{SEED}.

References

[1] Our paper

@inproceedings{xing2021kitchenshift,
    title={KitchenShift: Evaluating Zero-Shot Generalization of Imitation-Based Policy Learning Under Domain Shifts},
    author={Xing, Eliot and Gupta, Abhinav and Powers*, Sam and Dean*, Victoria},
    booktitle={NeurIPS 2021 Workshop on Distribution Shifts: Connecting Methods and Applications},
    year={2021},
    url={https://openreview.net/forum?id=DdglKo8hBq0}
}

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages