Adapt

September 19, 2023 · View on GitHub

Actuator Degeneration Adaptation Transformer. In submission progress of DAI 2023.

Videos

https://github.com/WentDong/Adapt/assets/36011596/14a5e2e0-196c-48e0-8838-38e19581c6c6

Environment Require

Nvidia IsaacGym
PyTorch

Details could be viewd in requirements.txt.

Teacher Policy

To train a teacher policy, please follow the example command:

python legged_gym/scripts/train.py --task a1_amp --sim_device $DEVICE --rl_device $DEVICE \
	--experiment_name $EXP --rum_name $RUN --max_iteration $ITER --joint $JOINT --seed $SEED

$JOINT means the id of joint whose actuator is suffering the degradation.
To evaluate, please follow the example command:

python legged_gym/scripts/evaluate.py --task a1_amp --sim_device $DEVICE --rl_device $DEVICE \
	--experiment_name $EXP --load_run $RUN --checkpoint $CHECKPOINT --file_name $FILE --joint $JOINT

If you want to test the performance over all 12 situations, please set $JOINT=-1.
To collect dataset, please follow the example command:

python legged_gym/scripts/collect.py --task a1_amp --sim_device $DEVICE --rl_device $DEVICE \
	--experiment_name $EXP --load_run $RUN --checkpoint $CHECKPOINT --file_name $FILE --joint $JOINT

To visualize the performance, please follow the example command:

python legged_gym/scripts/play.py --task a1_amp --sim_device $DEVICE --rl_device $DEVICE \
	--experiment_name $EXP --load_run $RUN --checkpoint $CHECKPOINT --joint $JOINT --rate $RATE

$RATE means the degradation rate, -1 for randomization.

Student Policy

To train a student policy, please follow the example command:
```
python scripts/train_Adapt.py
```
The detail args could be viewed in args.yaml.
To evaluate, please follow the example command:
```
python scripts/evaluate_Adapt.py
```
The detail args could be viewed in test_args.yaml.
To visualize the performance, please follow the example command:
```
python scripts/play_Adapt.py
```
The detail args could be viewed in test_args.yaml.