What question did this study set out to answer?

To analyze representation loss in offline hierarchical reinforcement learning and develop a mechanism to discover expressive, lossless action primitives.

June 26, 2023Open Access

Flow to Control: Offline Reinforcement Learning with Lossless Primitive Discovery

View Full Paper

Ask AI

Bookmark

Authors

YYYiqin Yang HHHao Hu WLWenzhe Li

Discussion

Member takes

Overview

Benchmark evaluation demonstrates superior hierarchical control in offline reinforcement learning, indicating that lossless primitives preserve essential policy expressiveness.

Key Points

To analyze representation loss in offline hierarchical reinforcement learning and develop a mechanism to discover expressive, lossless action primitives.
Formulated a quantitative analysis of policy expressiveness in offline hierarchical reinforcement learning.
Implemented a flow-based architecture for low-level policy representations to fully recover the original policy space from static datasets.
Evaluated agent performance against existing baselines using the standard D4RL benchmark suite alongside extensive ablation studies.
Flow-based primitive representation successfully captured behavioral datasets without restricting coverage of the overall policy space.
Achieved superior performance across the majority of evaluated D4RL benchmark tasks compared to prior hierarchical offline methods.

Cite This Study

Yang et al. (2023) studied this question.

synapsesocial.com/papers/6a104f7201be78fe8160b48e https://doi.org/10.1609/aaai.v37i9.26286

View Full Paper

Ask AI

Bookmark

June 26, 2023Open Access

Flow to Control: Offline Reinforcement Learning with Lossless Primitive Discovery

View Full Paper

Ask AI

Bookmark

Authors

YYYiqin Yang HHHao Hu WLWenzhe Li

Discussion

Member takes

Overview

Benchmark evaluation demonstrates superior hierarchical control in offline reinforcement learning, indicating that lossless primitives preserve essential policy expressiveness.

Key Points

To analyze representation loss in offline hierarchical reinforcement learning and develop a mechanism to discover expressive, lossless action primitives.
Formulated a quantitative analysis of policy expressiveness in offline hierarchical reinforcement learning.
Implemented a flow-based architecture for low-level policy representations to fully recover the original policy space from static datasets.
Evaluated agent performance against existing baselines using the standard D4RL benchmark suite alongside extensive ablation studies.
Flow-based primitive representation successfully captured behavioral datasets without restricting coverage of the overall policy space.
Achieved superior performance across the majority of evaluated D4RL benchmark tasks compared to prior hierarchical offline methods.

Cite This Study

Yang et al. (2023) studied this question.

synapsesocial.com/papers/6a104f7201be78fe8160b48e https://doi.org/10.1609/aaai.v37i9.26286

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark