A new method trains normalizing flows using constraint violation signals to map latent actions to feasible actions, reducing constraint violations by over 10x in several RL control benchmarks.
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
-
Leveraging Constraint Violation Signals For Action-Constrained Reinforcement Learning
A new method trains normalizing flows using constraint violation signals to map latent actions to feasible actions, reducing constraint violations by over 10x in several RL control benchmarks.