PACER: Process-Aware VLA Policy Learning from Human Corrections for Robotic Disassembly
PACER: Process-Aware VLA Policy Learning from Human Corrections for Robotic Disassembly
This paper presents PACER, a Process-Aware Correction and Evidence Reweighting framework for learning robotic disassembly policies from human corrections. The method improves policy performance by weighting valid action chunks based on progress toward the target and trace outcomes. Evaluated on UR5e motherboard disassembly, PACER achieves 87/100 successful approach-and-handoff trials, outperforming vanilla post-SFT methods with 54/100.
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
successful approach-and-handoff trials
87
View original evidence
87/100 successful approach-and-handoff trials on the held-out benchmark
Open source S4successful approach-and-handoff trials
54
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
and outcome to weight valid action chunks in the native supervised action loss. PACER trains one policy per weighting setting and selects one of the resulting policies through held-out validation. We evaluate PACER on UR5e motherboard disassembly across five component classes, where it achieves 87/100 successful approach-and-handoff trials on the held-out benchmark, compared with 54/100 for vanilla post-SFT.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org