A New Strategy for Verifying Reach-Avoid Specifications in Neural Feedback Systems
Samuel I. Akinwande, Sydney M. Katz, Mykel J. Kochenderfer, and Clark Barrett · 2026
Read A New Strategy for Verifying Reach-Avoid Specifications in Neural Feedback Systems by Samuel I. Akinwande, Sydney M. Katz, Mykel J. Kochenderfer, and Clark Barrett (2026) 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-2601-08065.txt.
Description
Samuel I. Akinwande and colleagues verify neural controlled systems by working backward from the goal. Reach-avoid specifications demand that a system reach a target while avoiding danger. Forward reachability analysis dominates verification of neural feedback systems. Backward methods have lagged because they scale poorly. Their 2026 preprint introduces algorithms computing both over- and under-approximations of backward reachable sets. Those sets collect states that can reach the goal safely.
Two-sided approximations bracket the true set from outside and inside. That bracket tells engineers how much trust a verdict deserves. Neural controllers in robots and vehicles need such checks before deployment. Backward reasoning starts from the goal, often a smaller set than the start region. The paper serves formal methods and control scholars. It strengthens a weaker side of the verification toolbox.
Catalog record
| CREATOR | Samuel I. Akinwande, Sydney M. Katz, Mykel J. Kochenderfer, and Clark Barrett |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 12 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-08065