Policy Contrastive Imitation Learning

Jialei Huang, Zhao-Heng Yin, Yingdong Hu, Yang Gao
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:14007-14022, 2023.

Abstract

Adversarial imitation learning (AIL) is a popular method that has recently achieved much success. However, the performance of AIL is still unsatisfactory on the more challenging tasks. We find that one of the major reasons is due to the low quality of AIL discriminator representation. Since the AIL discriminator is trained via binary classification that does not necessarily discriminate the policy from the expert in a meaningful way, the resulting reward might not be meaningful either. We propose a new method called Policy Contrastive Imitation Learning (PCIL) to resolve this issue. PCIL learns a contrastive representation space by anchoring on different policies and uses a smooth cosine-similarity-based reward to encourage imitation learning. Our proposed representation learning objective can be viewed as a stronger version of the AIL objective and provide a more meaningful comparison between the agent and the policy. From a theoretical perspective, we show the validity of our method using the apprenticeship learning framework. Furthermore, our empirical evaluation on the DeepMind Control suite demonstrates that PCIL can achieve state-of-the-art performance. Finally, qualitative results suggest that PCIL builds a smoother and more meaningful representation space for imitation learning.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-huang23n, title = {Policy Contrastive Imitation Learning}, author = {Huang, Jialei and Yin, Zhao-Heng and Hu, Yingdong and Gao, Yang}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {14007--14022}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/huang23n/huang23n.pdf}, url = {https://proceedings.mlr.press/v202/huang23n.html}, abstract = {Adversarial imitation learning (AIL) is a popular method that has recently achieved much success. However, the performance of AIL is still unsatisfactory on the more challenging tasks. We find that one of the major reasons is due to the low quality of AIL discriminator representation. Since the AIL discriminator is trained via binary classification that does not necessarily discriminate the policy from the expert in a meaningful way, the resulting reward might not be meaningful either. We propose a new method called Policy Contrastive Imitation Learning (PCIL) to resolve this issue. PCIL learns a contrastive representation space by anchoring on different policies and uses a smooth cosine-similarity-based reward to encourage imitation learning. Our proposed representation learning objective can be viewed as a stronger version of the AIL objective and provide a more meaningful comparison between the agent and the policy. From a theoretical perspective, we show the validity of our method using the apprenticeship learning framework. Furthermore, our empirical evaluation on the DeepMind Control suite demonstrates that PCIL can achieve state-of-the-art performance. Finally, qualitative results suggest that PCIL builds a smoother and more meaningful representation space for imitation learning.} }
Endnote
%0 Conference Paper %T Policy Contrastive Imitation Learning %A Jialei Huang %A Zhao-Heng Yin %A Yingdong Hu %A Yang Gao %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-huang23n %I PMLR %P 14007--14022 %U https://proceedings.mlr.press/v202/huang23n.html %V 202 %X Adversarial imitation learning (AIL) is a popular method that has recently achieved much success. However, the performance of AIL is still unsatisfactory on the more challenging tasks. We find that one of the major reasons is due to the low quality of AIL discriminator representation. Since the AIL discriminator is trained via binary classification that does not necessarily discriminate the policy from the expert in a meaningful way, the resulting reward might not be meaningful either. We propose a new method called Policy Contrastive Imitation Learning (PCIL) to resolve this issue. PCIL learns a contrastive representation space by anchoring on different policies and uses a smooth cosine-similarity-based reward to encourage imitation learning. Our proposed representation learning objective can be viewed as a stronger version of the AIL objective and provide a more meaningful comparison between the agent and the policy. From a theoretical perspective, we show the validity of our method using the apprenticeship learning framework. Furthermore, our empirical evaluation on the DeepMind Control suite demonstrates that PCIL can achieve state-of-the-art performance. Finally, qualitative results suggest that PCIL builds a smoother and more meaningful representation space for imitation learning.
APA
Huang, J., Yin, Z., Hu, Y. & Gao, Y.. (2023). Policy Contrastive Imitation Learning. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:14007-14022 Available from https://proceedings.mlr.press/v202/huang23n.html.

Related Material