Hierarchical Imitation Learning with Vector Quantized Models

Kalle Kujanpää, Joni Pajarinen, Alexander Ilin
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:17896-17919, 2023.

Abstract

The ability to plan actions on multiple levels of abstraction enables intelligent agents to solve complex tasks effectively. However, learning the models for both low and high-level planning from demonstrations has proven challenging, especially with higher-dimensional inputs. To address this issue, we propose to use reinforcement learning to identify subgoals in expert trajectories by associating the magnitude of the rewards with the predictability of low-level actions given the state and the chosen subgoal. We build a vector-quantized generative model for the identified subgoals to perform subgoal-level planning. In experiments, the algorithm excels at solving complex, long-horizon decision-making problems outperforming state-of-the-art. Because of its ability to plan, our algorithm can find better trajectories than the ones in the training set.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-kujanpaa23a, title = {Hierarchical Imitation Learning with Vector Quantized Models}, author = {Kujanp\"{a}\"{a}, Kalle and Pajarinen, Joni and Ilin, Alexander}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {17896--17919}, 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/kujanpaa23a/kujanpaa23a.pdf}, url = {https://proceedings.mlr.press/v202/kujanpaa23a.html}, abstract = {The ability to plan actions on multiple levels of abstraction enables intelligent agents to solve complex tasks effectively. However, learning the models for both low and high-level planning from demonstrations has proven challenging, especially with higher-dimensional inputs. To address this issue, we propose to use reinforcement learning to identify subgoals in expert trajectories by associating the magnitude of the rewards with the predictability of low-level actions given the state and the chosen subgoal. We build a vector-quantized generative model for the identified subgoals to perform subgoal-level planning. In experiments, the algorithm excels at solving complex, long-horizon decision-making problems outperforming state-of-the-art. Because of its ability to plan, our algorithm can find better trajectories than the ones in the training set.} }
Endnote
%0 Conference Paper %T Hierarchical Imitation Learning with Vector Quantized Models %A Kalle Kujanpää %A Joni Pajarinen %A Alexander Ilin %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-kujanpaa23a %I PMLR %P 17896--17919 %U https://proceedings.mlr.press/v202/kujanpaa23a.html %V 202 %X The ability to plan actions on multiple levels of abstraction enables intelligent agents to solve complex tasks effectively. However, learning the models for both low and high-level planning from demonstrations has proven challenging, especially with higher-dimensional inputs. To address this issue, we propose to use reinforcement learning to identify subgoals in expert trajectories by associating the magnitude of the rewards with the predictability of low-level actions given the state and the chosen subgoal. We build a vector-quantized generative model for the identified subgoals to perform subgoal-level planning. In experiments, the algorithm excels at solving complex, long-horizon decision-making problems outperforming state-of-the-art. Because of its ability to plan, our algorithm can find better trajectories than the ones in the training set.
APA
Kujanpää, K., Pajarinen, J. & Ilin, A.. (2023). Hierarchical Imitation Learning with Vector Quantized Models. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:17896-17919 Available from https://proceedings.mlr.press/v202/kujanpaa23a.html.

Related Material