From Demonstrations to Safe Deployment: Path-Consistent Safety Filtering for Diffusion Policies

2025/11/09 by Römer, Ralf, Balletshofer, Julian, Thumm, Jakob +3
Computer Science · Engineering · #Adversarial Robustness in Machine Learning #Control (management) #FOS: Computer and information sciences #FOS: Electrical engineering #I.2.10 #I.2.6 #I.2.8 #I.2.9 #Key (lock) #Reachability #Reinforcement Learning in Robotics #Robot Manipulation and Learning #Robotics (cs.RO) #Sequence (biology) #Software deployment #System safety #Systems and Control (eess.SY) #Task (project management) #Trajectory #electronic engineering #information engineering

paper · open access · doi:10.48550/arxiv.2511.06385

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/11/09 · openalex created_date 2025/11/12 · openalex updated_date 2026/07/28

Abstract

Diffusion policies (DPs) achieve state-of-the-art performance on complex manipulation tasks by learning from large-scale demonstration datasets, often spanning multiple embodiments and environments. However, they cannot guarantee safe behavior, requiring external safety mechanisms. These, however, alter actions in ways unseen during training, causing unpredictable behavior and performance degradation. To address these problems, we propose path-consistent safety filtering (PACS) for DPs. Our approach performs path-consistent braking on a trajectory computed from the sequence of generated actions. In this way, we keep the execution consistent with the training distribution of the policy, maintaining the learned, task-completing behavior. To enable real-time deployment and handle uncertainties, we verify safety using set-based reachability analysis. Our experimental evaluation in simulation and on three challenging real-world human-robot interaction tasks shows that PACS (a) provides formal safety guarantees in dynamic environments, (b) preserves task success rates, and (c) outperforms reactive safety approaches, such as control barrier functions, by up to 68 % in terms of task success. Videos are available at our project website: https://tum-lsy.github.io/pacs.

Citations

• SAD-Flower: Flow Matching for Safe, Admissible, and Dynamically Consistent Planning
• Failure Prediction at Runtime for Generative Robot Policies
• CRISP -- Compliant ROS2 Controllers for Learning-Based Manipulation Policies and Teleoperation
• SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics
• Can We Detect Failures Without Failure Data? Uncertainty-Aware Runtime Failure Detection for Imitation Learning Policies
• Safe, Task-Consistent Manipulation with Operational Space Control Barrier Functions
• RA-DP: Rapid Adaptive Diffusion Policy for Training-Free High-frequency Robotics Replanning
• Simultaneous Multi-Robot Motion Planning with Projected Diffusion Models
• Motion Planning Diffusion: Learning and Adapting Robot Motion Planning with Diffusion Models
• A General Safety Framework for Autonomous Manipulation in Human Environments
• Diffusion Predictive Control with Constraints
• π₀: A Vision-Language-Action Flow Model for General Robot Control
• One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation
• Denoising Diffusion Planner: Learning Complex Paths from Low-Quality Demonstrations
• Unpacking Failure Modes of Generative Policies: Runtime Monitoring of Consistency and Progress
• RAIL: Reachability-Aided Imitation Learning for Safe Policy Execution
• Towards Unconstrained Collision Injury Protection Data Sets: Initial Surrogate Experiments for the Human Hand
• Multimodal Diffusion Transformer: Learning Versatile Behavior from Multimodal Goals
• Towards Safe Robot Use with Edged or Pointed Objects: A Surrogate Study Assembling a Human Hand Injury Protection Database
• Human-Robot Gym: Benchmarking Reinforcement Learning in Human-Robot Collaboration
• The Safety Filter: A Unified View of Safety-Critical Control in Autonomous Systems
• SafeDiffuser: Safe Planning with Diffusion Probabilistic Models
• Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
• Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
• Diffusion policy: Visuomotor policy learning via action diffusion
• Flow Matching for Generative Modeling
• Provably Safe Deep Reinforcement Learning for Robotic Manipulation in Human Environments
• Safety-Critical Manipulation for Collision-Free Food Preparation
• Safe Learning in Robotics: From Learning-Based Control to Safe Reinforcement Learning
• What Matters in Learning from Offline Human Demonstrations for Robot Manipulation
• Jerk-limited Real-time Trajectory Generation with Arbitrary Target States
• A predictive safety filter for learning-based racing control
• Reachability-based Trajectory Safeguard (RTS): A Safe and Fast Reinforcement Learning Safety Layer for Continuous Control
• Denoising Diffusion Probabilistic Models
• Safe Reinforcement Learning via Projection on a Safe Set: How to Achieve Optimality?
• Control Barrier Functions: Theory and Applications
• A predictive safety filter for learning-based control of constrained\n nonlinear dynamical systems
• A General Safety Framework for Learning-Based Control in Uncertain\n Robotic Systems

Related

• SafeDiffuser: Safe Planning with Diffusion Probabilistic Models
• CBF-RL: Safety Filtering Reinforcement Learning in Training with Control Barrier Functions
• SafeFlow: Safe Robot Motion Planning with Flow Matching via Control Barrier Functions
• SafeBimanual: Diffusion-based Trajectory Optimization for Safe Bimanual Manipulation
• Safe Deep Policy Adaptation