Statistically Assuring Safety of Control Systems using Ensembles of Safety Filters and Conformal Prediction

2025/11/11 by Ihab Tabbara, Tabbara, Ihab, Yuxuan Yang +3
Computer Science · Engineering · #Adversarial Robustness in Machine Learning #Artificial Intelligence (cs.AI) #Bellman equation #FOS: Computer and information sciences #FOS: Electrical engineering #Filter (signal processing) #Formal Methods in Verification #Function (biology) #Leverage (statistics) #Machine Learning (cs.LG) #Probabilistic logic #Reachability #Reinforcement learning #Robotics (cs.RO) #Set (abstract data type) #Smart Grid Security and Resilience #System safety #Systems and Control (eess.SY) #electronic engineering #information engineering

paper · pdf · doi:10.48550/arxiv.2511.07899

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/11/11 · openalex created_date 2025/11/13 · openalex updated_date 2026/07/28

Abstract

Safety assurance is a fundamental requirement for deploying learning-enabled autonomous systems. Hamilton-Jacobi (HJ) reachability analysis is a fundamental method for formally verifying safety and generating safe controllers. However, computing the HJ value function that characterizes the backward reachable set (BRS) of a set of user-defined failure states is computationally expensive, especially for high-dimensional systems, motivating the use of reinforcement learning approaches to approximate the value function. Unfortunately, a learned value function and its corresponding safe policy are not guaranteed to be correct. The learned value function evaluated at a given state may not be equal to the actual safety return achieved by following the learned safe policy. To address this challenge, we introduce a conformal prediction-based (CP) framework that bounds such uncertainty. We leverage CP to provide probabilistic safety guarantees when using learned HJ value functions and policies to prevent control systems from reaching failure states. Specifically, we use CP to calibrate the switching between the unsafe nominal controller and the learned HJ-based safe policy and to derive safety guarantees under this switched policy. We also investigate using an ensemble of independently trained HJ value functions as a safety filter and compare this ensemble approach to using individual value functions alone.

Citations

• Designing Latent Safety Filters using Pre-Trained Vision Models
• Reachability Barrier Networks: Learning Hamilton-Jacobi Solutions for Smooth and Flexible Control Barrier Functions
• Uncertainty-aware Latent Safety Filters for Avoiding Out-of-Distribution Failures
• Generalizing Safety Beyond Collision-Avoidance via Latent-Space Reachability Analysis
• Safe Decentralized Multi-Agent Control using Black-Box Predictors, Conformal Decision Policies, and Control Barrier Functions
• Formal Verification and Control with Conformal Prediction
• Certifiable Reachability Learning Using a New Lipschitz Continuous Value Function
• Verification of Neural Reachable Tubes via Scenario Optimization and Conformal Prediction
• End-to-end Autonomous Driving: Challenges and Frontiers
• Safe Planning in Dynamic Environments using Conformal Prediction
• A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification
• DeepReach: A Deep Learning Approach to High-Dimensional Reachability
• Hamilton-Jacobi Reachability: A Brief Overview and Recent Advances
• Continuous control with deep reinforcement learning

Related

• A Physics-Informed Machine Learning Framework for Safe and Optimal Control of Autonomous Systems
• Hamilton-Jacobi Reachability in Reinforcement Learning: A Survey
• Patching Approximately Safe Value Functions Leveraging Local Hamilton-Jacobi Reachability Analysis
• A General Safety Framework for Learning-Based Control in Uncertain\n Robotic Systems
• Computationally and Sample Efficient Safe Reinforcement Learning Using Adaptive Conformal Prediction