Learning Neural Control Barrier Functions from Offline Data with Conservatism

2025-05-01

AI Summary

Control Barrier Function (CBF)-based safety filters suffer from the curse of dimensionality and poor out-of-distribution (OOD) robustness in high-dimensional systems. Method: This paper proposes Conservative Control Barrier Functions (CCBFs), the first framework to integrate conservative reinforcement learning principles into CBF learning. We parameterize the CBF using a deep neural network and formulate an offline conservative optimization objective that jointly incorporates safety-set constraint regularization and an OOD state rejection mechanism. Contribution/Results: Compared to existing online and offline CBF methods, CCBFs achieve significantly improved safety guarantees and OOD detection capability across multiple nonlinear dynamical tasks, while maintaining near-optimal task performance. The approach simultaneously ensures satisfaction of safety constraints and robustness to distributional shift—addressing two critical limitations of prior CBF-based safety filters.

Technology Category

• Natural Language Processing: Safety and Robustness
• Intelligent Robots: Learning & Optimization for ROB
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Safety filters, particularly those based on control barrier functions, have gained increased interest as effective tools for safe control of dynamical systems. Existing correct-by-construction synthesis algorithms, however, suffer from the curse of dimensionality. Deep learning approaches have been proposed in recent years to address this challenge. In this paper, we contribute to this line of work by proposing an algorithm for training control barrier functions from offline datasets. Our algorithm trains the filter to not only prevent the system from reaching unsafe states but also out-of-distribution ones, at which the filter would be unreliable. It is inspired by Conservative Q-learning, an offline reinforcement learning algorithm. We call its outputs Conservative Control Barrier Functions (CCBFs). Our empirical results demonstrate that CCBFs outperform existing methods in maintaining safety and out-of-distribution avoidance while minimally affecting task performance.

Problem

• Learning safe control barrier functions from offline data
• Preventing system from reaching unsafe and unreliable states
• Improving safety and out-of-distribution avoidance performance

Innovation

• Learning control barrier functions from offline data
• Preventing unsafe and out-of-distribution states
• Inspired by Conservative Q-learning algorithm