Training a VLM planner on recovery traces synthesized from perturbations and privileged corrections lifts dynamic household task success from 33.3% to 76.2%.
Affordance-R1: Reinforcement Learning for Generalizable Affordance Reasoning in Multimodal Large Language Models , booktitle =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
From Failures to Supervision: DynamicEnvPlan for Robust Long-Horizon Embodied Planning
Training a VLM planner on recovery traces synthesized from perturbations and privileged corrections lifts dynamic household task success from 33.3% to 76.2%.