Eventual Discounting Temporal Logic Counterfactual Experience Replay

Cameron Voloshin, Abhinav Verma, Yisong Yue
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:35137-35150, 2023.

Abstract

Linear temporal logic (LTL) offers a simplified way of specifying tasks for policy optimization that may otherwise be difficult to describe with scalar reward functions. However, the standard RL framework can be too myopic to find maximally LTL satisfying policies. This paper makes two contributions. First, we develop a new value-function based proxy, using a technique we call eventual discounting, under which one can find policies that satisfy the LTL specification with highest achievable probability. Second, we develop a new experience replay method for generating off-policy data from on-policy rollouts via counterfactual reasoning on different ways of satisfying the LTL specification. Our experiments, conducted in both discrete and continuous state-action spaces, confirm the effectiveness of our counterfactual experience replay approach.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-voloshin23a, title = {Eventual Discounting Temporal Logic Counterfactual Experience Replay}, author = {Voloshin, Cameron and Verma, Abhinav and Yue, Yisong}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {35137--35150}, 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/voloshin23a/voloshin23a.pdf}, url = {https://proceedings.mlr.press/v202/voloshin23a.html}, abstract = {Linear temporal logic (LTL) offers a simplified way of specifying tasks for policy optimization that may otherwise be difficult to describe with scalar reward functions. However, the standard RL framework can be too myopic to find maximally LTL satisfying policies. This paper makes two contributions. First, we develop a new value-function based proxy, using a technique we call eventual discounting, under which one can find policies that satisfy the LTL specification with highest achievable probability. Second, we develop a new experience replay method for generating off-policy data from on-policy rollouts via counterfactual reasoning on different ways of satisfying the LTL specification. Our experiments, conducted in both discrete and continuous state-action spaces, confirm the effectiveness of our counterfactual experience replay approach.} }
Endnote
%0 Conference Paper %T Eventual Discounting Temporal Logic Counterfactual Experience Replay %A Cameron Voloshin %A Abhinav Verma %A Yisong Yue %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-voloshin23a %I PMLR %P 35137--35150 %U https://proceedings.mlr.press/v202/voloshin23a.html %V 202 %X Linear temporal logic (LTL) offers a simplified way of specifying tasks for policy optimization that may otherwise be difficult to describe with scalar reward functions. However, the standard RL framework can be too myopic to find maximally LTL satisfying policies. This paper makes two contributions. First, we develop a new value-function based proxy, using a technique we call eventual discounting, under which one can find policies that satisfy the LTL specification with highest achievable probability. Second, we develop a new experience replay method for generating off-policy data from on-policy rollouts via counterfactual reasoning on different ways of satisfying the LTL specification. Our experiments, conducted in both discrete and continuous state-action spaces, confirm the effectiveness of our counterfactual experience replay approach.
APA
Voloshin, C., Verma, A. & Yue, Y.. (2023). Eventual Discounting Temporal Logic Counterfactual Experience Replay. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:35137-35150 Available from https://proceedings.mlr.press/v202/voloshin23a.html.

Related Material