Training a diverse set of policies, relabeling their experience with the target reward, and selecting the highest-return buffer improves model-based offline RL on unknown tasks, supported by a Wasserstein-distance analysis.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Unsupervised Data Generation for Offline Reinforcement Learning: A Perspective from Model
Training a diverse set of policies, relabeling their experience with the target reward, and selecting the highest-return buffer improves model-based offline RL on unknown tasks, supported by a Wasserstein-distance analysis.