robotics
The Robot Remembers How Close It Was
Value as Memory feeds recent value predictions back into a general policy, then updates it during real-world practice.
Summary
Value as Memory feeds recent value predictions back into a general policy, then updates it during real-world practice.
Partial observability can leave a robot unsure whether a long task is advancing or repeating. Value as Memory caches recent value predictions as a compact progress signal and combines offline demonstrations, human interventions and autonomous rollouts for online reinforcement learning. On the authors' deformable-object and contact-rich tasks, value caching produced a 29.4-point mean absolute success-rate gain, while online learning added 32.8 points over matched offline training. Concurrent collection and training also completed eight times as many policy iterations under the same gradient-step budget. These results cover the reported tasks and hardware, not general household autonomy.
Why it matters
Value as Memory feeds recent value predictions back into a general policy, then updates it during real-world practice.
Limits and context
- These results cover the reported tasks and hardware, not general household autonomy.
Key claims
Value as Memory feeds recent value predictions back into a general policy, then updates it during real-world practice.
Qualification: These results cover the reported tasks and hardware, not general household autonomy.
Evidence: source-2026-10-17-003
Sources
- arXiv preprint 2610.22368, original v1arXiv · primary research
Corrections
No corrections have been recorded for this story.