Confidence-Modulated Safety Filtering for End-to-End Policies
Confidence-Modulated Safety Filtering for End-to-End Policies
This research paper introduces a method where the conservatism of a safety filter is dynamically adjusted based on the confidence level of an end-to-end robot policy. The approach aims to maximize performance within the filter's certified operational domain while adding necessary safety margins when the policy operates beyond its competence or flags potential failures, validated through preliminary simulations on the F1TENTH platform.
The authors propose a runtime confidence-modulation mechanism for safety filters applied to end-to-end robot policies, aiming to balance performance within the filter's operational design domain with additional safety margins when policy confidence is low or outside that domain; preliminary validation is limited to F1TENTH simulations.
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.
- Environment
- SimulationOpen source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
dds margin, which helps when it flags where the filter's model fails. Preliminary F1TENTH simulations, where this holds, show the modulated filter combining the strengths of permissive and conservative filters.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org