ORN-CBF: Learning Observation-conditioned Residual Neural Control Barrier Functions via Hypernetworks

2025-09-20

AI Summary

Designing control barrier functions (CBFs) under partial observability remains challenging due to difficulties in constructing provably safe sets, suboptimal safety set design, and lack of rigorous safety guarantees. Method: This paper proposes an observation-conditioned neural CBF framework grounded in Hamilton–Jacobi (HJ) reachability analysis. Leveraging the mathematical properties of the HJ value function, the method ensures that the learned safe set strictly avoids the failure set. A hypernetwork architecture jointly models state estimation and residual CBF synthesis—the first application of hypernetworks to observation-driven residual neural CBF design. The maximal safe set is approximated via deep neural networks. Evaluation is conducted on ground robots and quadrotors in both simulation and real hardware. Results: Our approach significantly improves task success rates over baseline methods and demonstrates strong out-of-distribution generalization capability.

Technology Category

• Intelligent Robots: State Estimation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Natural Language Processing: Safety and Robustness

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Control barrier functions (CBFs) have been demonstrated as an effective method for safety-critical control of autonomous systems. Although CBFs are simple to deploy, their design remains challenging, motivating the development of learning-based approaches. Yet, issues such as suboptimal safe sets, applicability in partially observable environments, and lack of rigorous safety guarantees persist. In this work, we propose observation-conditioned neural CBFs based on Hamilton-Jacobi (HJ) reachability analysis, which approximately recover the maximal safe sets. We exploit certain mathematical properties of the HJ value function, ensuring that the predicted safe set never intersects with the observed failure set. Moreover, we leverage a hypernetwork-based architecture that is particularly suitable for the design of observation-conditioned safety filters. The proposed method is examined both in simulation and hardware experiments for a ground robot and a quadcopter. The results show improved success rates and generalization to out-of-domain environments compared to the baselines.

Problem

• Learning safety-critical control for autonomous systems in partially observable environments
• Addressing suboptimal safe sets and lack of rigorous safety guarantees
• Developing observation-conditioned safety filters with improved generalization capabilities

Innovation

• Observation-conditioned neural CBFs based on Hamilton-Jacobi reachability
• Hypernetwork architecture for observation-conditioned safety filters
• Mathematical HJ value function properties ensure safe set separation