Neural Integral Control Barrier Functions for Agile Collision Avoidance
Neural Integral Control Barrier Functions for Agile Collision Avoidance
This paper introduces a learning-based approach to integral control barrier functions (ICBFs) that jointly learns dynamics, nominal controllers, and barrier functions from data. By training the barrier on viability rather than geometry and incorporating actuator limits as hard bounds, the method ensures safety even under model errors. Experiments on a 6-D dynamic bicycle demonstrate that the proposed filter removes all 12 unsafe episodes from 200 held-out layouts while operating above 100 Hz on a single CPU core, outperforming a jointly trained baseline in constraint adherence.
This preprint proposes a data-driven framework using neural ODEs and imitation learning to construct integral control barrier functions for agile collision avoidance, validated on a 6-D dynamic bicycle model where the method eliminates all unsafe episodes in held-out layouts while maintaining high-frequency solvability.
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.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
he input it returns can drift from the one the true models would give. On agile racing with a 6-D dynamic bicycle, the filter removes all 12 unsafe episodes from 200 held out layouts at an unchanged success rate and above 100 Hz on one CPU core, holds the input constraint on every episode of a benchmark where a jointly trained baseline holds it on 53.38%, and keeps that advantage under large errors in the plant frict
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org