Safe Execution of RL Policies via Acceleration-based CBF-QP Constraint Enforcement for Real-World Robotic Deployments
This research paper introduces Acc-CBF-QP, an acceleration-based Quadratic Program safety filter utilizing Control Barrier Functions to constrain reinforcement learning policies onto a safe set at runtime without requiring policy retraining. The method enforces joint position, velocity, torque, and collision constraints within a unified optimization framework and incorporates disturbance observers to mitigate modeling errors. Experiments on a 7-DoF Kinova Gen3 manipulator and a 19-DoF Unitree H1 humanoid demonstrate substantial reductions in constraint violations and hardware-damaging failures while preserving nominal task performance.
Summary and content fields are drawn from the document text and are pending editorial review.
Who issued it
Who it governs
Robotics researchers and engineers looking to deploy learned reinforcement learning policies safely on physical hardware.
What the document contains
The document addresses the safe execution of reinforcement learning policies on physical robotic hardware by interposing an acceleration-based Quadratic Program safety filter with Control Barrier Functions. It covers dynamic constraint enforcement for joint limits, velocities, torques, and collisions, along with external disturbance estimation using a generalized-momentum observer.
The document does not provide legislative or regulatory governance guidelines, policy frameworks for commercial deployment outside of robotics, or formal closed-loop stability proofs.
- Acc-CBF-QP serves as a runtime safety filter that constrains reinforcement learning policies onto a safe set without modifying training.
- The framework unifies joint position, velocity, torque, and collision constraints using an acceleration-based Quadratic Program.
- Experimental validation on a Kinova Gen3 manipulator and a Unitree H1 humanoid demonstrates major reductions in constraint violations.
- The TorqueTask formulation exhibits greater robustness to modeling errors compared to the Forward Dynamics Task.
- Safety enforcement is achieved without degrading nominal task performance in violation-free regimes.
· I. INTRODUCTION · II. RELATED WORKS · III. METHOD · IV. EXPERIMENTS · V. LIMITATIONS & FUTURE WORK · VI. CONCLUSION
How it scores
Each score is backed by a quote verified against the document text. Editorial review has not yet confirmed them.
Accountability
0/5Clear responsibility for AI outcomes
screening score · 2026-10-23
Fairness
0/5Non-discrimination and equitable treatment
screening score · 2026-10-23
Human Dignity
0/5Respect for the inherent worth of persons
screening score · 2026-10-23
Human Oversight
0/5Humans remain in control of consequential decisions
screening score · 2026-10-23
Privacy
0/5Protection of personal data and private life
screening score · 2026-10-23
Transparency
4/5Openness about how AI systems work and are used
The full pipeline is open-source.
I. INTRODUCTION
screening score · 2026-10-23