Ontology-Grounded World Models for Failure Diagnosis and Closed-Loop Repair in Physical AI Systems
2.80T1 sourcearXiv cs.RO
Source record
Published by arXiv cs.RO (T1 source). The original is at https://arxiv.org/abs/2608.13901.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryPresents Onto-EV-WM, an ontology-grounded interface layered above an existing world model (EV-WM) for diagnosing failures and gating corrections in physical AI systems. Reports benchmark results on PointMaze and LIBERO-Goal/Plus with success rates of 85–94% depending on configuration.
Why it mattersDetailed benchmark numbers and a concrete symbolic-verification layer over a learned world model, but the contribution is incremental and the domain (robotic manipulation benchmarks) is far from everyday desk workflows.
Cited by
No citations on record.