New paper proposes method to measure conservatism of latent safety filters for world models
A new arXiv paper examines latent safety filters, which decide from a world model's latent whether a planned action is safe, and measures their conservatism. Conformal calibration is used to set the filter's threshold.
Read the full story at arXiv cs.RO ↗
Timeline · 1 report
- 2026-11-19 05:00 · arXiv cs.RO
Measuring the Conservatism of Latent Safety Filters for World Models