In simulation, ground-truth mask goal conditioning reaches about 90% grasp success in and out of distribution, but the proposed GroundingDINO-based pipeline fails to reproduce this when trained with detector-generated masks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Versatile and Generalizable Manipulation via Goal-Conditioned Reinforcement Learning with Grounded Object Detection
In simulation, ground-truth mask goal conditioning reaches about 90% grasp success in and out of distribution, but the proposed GroundingDINO-based pipeline fails to reproduce this when trained with detector-generated masks.