Generalizing Safety Beyond Collision-Avoidance via Latent-Space Reachability Analysis
Kensuke Nakamura, Lasse Peters, Andrea Bajcsy · 2025
Read Generalizing Safety Beyond Collision-Avoidance via Latent-Space Reachability Analysis by Kensuke Nakamura, Lasse Peters, Andrea Bajcsy (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-00935.txt.
Description
Robot safety checks usually guard against crashes, but some dangers are hard to write down, such as spilling the contents of a bag. Kensuke Nakamura, Lasse Peters, and Andrea Bajcsy build safety filters that work from raw camera images instead of hand-written rules. Their method learns a model of the world from mixed robot data, good and bad, then marks safe and unsafe spots inside that model's inner picture of each scene.
In tests on real and simulated robots, the filters guarded other controllers, from learned policies to a human driving by hand. They kept a robot arm from spilling a bag or knocking over clutter. The approach extends a strict math framework for safety to hazards seen, not stated.
Catalog record
| CREATOR | Kensuke Nakamura, Lasse Peters, Andrea Bajcsy |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-00935