Learning Neural Control Barrier Functions from Expert Demonstrations Using Inverse Constraint Learning
Yuxuan Yang, Hussein Sibai
AI summary
Problem
Training neural control barrier functions typically requires labeled safe and unsafe state datasets, but failure sets are often non-obvious or mathematically intractable to specify. Existing data-driven methods struggle with inefficient sampling and distribution shift in high-dimensional spaces.
Approach
The method uses inverse constraint learning to infer a constraint function that classifies states as safe or unsafe based on expert demonstrations. This constraint labels simulated trajectories to train a neural CBF in an iterative loop, with a heuristic that delays CBF training to accelerate convergence.
Key results
- Outperforms ROCBF and iDBF baselines across four robotic environments
- Matches performance of neural CBFs trained with ground-truth safety labels
- Introduces a training heuristic that delays CBF updates to accelerate convergence
- Successfully generalizes to multi-task expert demonstrations
Why it matters
Enables scalable, data-driven safety filter synthesis for autonomous systems where explicit failure sets are unknown, advancing safe deployment in robotics and autonomous control.
Abstract
Safety is a fundamental requirement for au- tonomous systems operating in critical domains. Control barrier functions (CBFs) have been used to design safety filters that minimally alter nominal controls for such systems to maintain their safety. Learning neural CBFs has been proposed as a data-driven alternative for their computationally expensive optimization-based synthesis. However, it is often the case that the failure set of states that should be avoided is non-obvious or hard to specify formally, e.g., tailgating in autonomous driving, while a set of expert demonstrations that achieve the task and avoid the failure set is easier to generate. We use inverse constraint learning (ICL) to train a constraint function that classifies the states of the system under consideration to safe, i.e., belong to a controlled forward invariant set that is disjoint from the unspecified failure set, and unsafe ones, i.e., belong to the complement of that set. We then use that function to label a new set of simulated trajectories to train our neural CBF. We empirically evaluate our approach in four different environments, demonstrating that it outperforms existing baselines and achieves comparable performance to a neural CBF trained with the same data but annotated with ground-truth safety labels.