When Should a Robot Retry a Skill? Evidence from Fault-Injected Humanoid Simulation
After a timeout, retrying can repeat a skill and advancing can skip it. We tested recovery in fault-injected humanoid simulation with a start deadline. With receipts, skill-aware and template rules each completed more tasks at one fault rate. Compared with receipts and inertial phase evidence, issued frames increased ordered completions from 26 to 43 at the lower fault rate and 10 to 20 at the higher (50 tasks each). Two policies given identical frames agreed. Program scoring credited 20 normal and 40 restrained waves. Shoulder excursion, elbow cycles and return posture separated these outcomes in 20 paired scenes. Author-submitted manuscript. This preprint has not been peer reviewed. The supplementary archive contains the simulation records, joint trajectories, and scoring code used in this study.
Authors
- Hongda Yan (ORCID: https://orcid.org/0009-0008-0235-0641)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-18
- DOI
- https://doi.org/10.5281/zenodo.24331687
- Primary Topic
- Robot Manipulation and Learning
- Type
- preprint