A model-free visual RL agent that jointly trains latent self-prediction and next-observation prediction, mediated by two adapters, improves aggregate DMControl scores over prior methods.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Observation-Grounded Self-Predictive Reinforcement Learning for Visual Continuous Control
A model-free visual RL agent that jointly trains latent self-prediction and next-observation prediction, mediated by two adapters, improves aggregate DMControl scores over prior methods.