REMM: Revisable Embodied Memory for Manipulation
REMM: Revisable Embodied Memory for Manipulation
This research proposes REMM, a supervisory memory layer for manipulation tasks that enables robots to retain and revise task-relevant history without retraining policies. By separating evidence history from task state and linking conclusions to supporting facts, REMM allows for targeted recovery after disturbances. Experimental results across 51 physical trials show improvements in task progress and sequence recall compared to baselines.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
physical trials
51
View original evidence
mory layer for a frozen G0.5 vision-language-action executor. REMM separates an append-only evidence history from revisable task state and explicitly links derived relations and completed conditions to their supporting facts. When accepted evidence changes a supporting fact, it withdraws dependent conclusions and schedules targeted recovery at verified skill boundaries without updating the executor. Across 51 physica
Open source S3
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
mory layer for a frozen G0.5 vision-language-action executor. REMM separates an append-only evidence history from revisable task state and explicitly links derived relations and completed conditions to their supporting facts. When accepted evidence changes a supporting fact, it withdraws dependent conclusions and schedules targeted recovery at verified skill boundaries without updating the executor. Across 51 physica
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org