Detecting Silent Action-Interface Mismatch with Calibrated Relational World Models
Detecting Silent Action-Interface Mismatch with Calibrated Relational World Models
This research paper introduces a method to detect silent action-interface mismatches in robotics by using learned state-action-effect relations as offline contracts. The authors evaluate their approach on ten LIBERO Spatial tasks, demonstrating that a four-component joint Gaussian mixture model achieves a 65.7% recall rate with a 95.8% localization score under a 2.5% false-positive rate. The study highlights the tradeoff between detection and localization and notes that earlier results on Object/Goal tasks do not extend to other benchmarks like DROID or HYDRA, where performance fails specific gates.
The study presents offline diagnostic methods for detecting semantic mismatches in robot actions using calibrated relational world models, evaluated on ten LIBERO Spatial tasks with specific recall and localization metrics that are explicitly described as retrospective diagnostics rather than deployment guarantees.
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
source episodes
432
Reported trials
View original evidence
ulary supports closed-set localization. On ten LIBERO Spatial tasks comprising 432 source episodes, a capacity-controlled comparison at 2.5% empirical clean false-positive rate gives 46.1% recall for a ridge forward residual, 61.0% for a four-component joint Gaussian mixture, and 65.7% for the exact conditional of that same mixture. The joint score retains a 14.9 percentage-point advantage over ridge, while condition
Open source S3Reported recall
46.1%
View original evidence
ulary supports closed-set localization. On ten LIBERO Spatial tasks comprising 432 source episodes, a capacity-controlled comparison at 2.5% empirical clean false-positive rate gives 46.1% recall for a ridge forward residual, 61.0% for a four-component joint Gaussian mixture, and 65.7% for the exact conditional of that same mixture. The joint score retains a 14.9 percentage-point advantage over ridge, while condition
Open source S3
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ulary supports closed-set localization. On ten LIBERO Spatial tasks comprising 432 source episodes, a capacity-controlled comparison at 2.5% empirical clean false-positive rate gives 46.1% recall for a ridge forward residual, 61.0% for a four-component joint Gaussian mixture, and 65.7% for the exact conditional of that same mixture. The joint score retains a 14.9 percentage-point advantage over ridge, while condition
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org