CBF-RL: Safety Filtering Reinforcement Learning in Training with Control Barrier Functions
Lizhi Yang, Blake Werner, Massimiliano de Sa, Aaron Ames
AI summary
Problem
Reinforcement learning often prioritizes performance over safety, leading to catastrophic failures in real-world deployments, while existing safety-filtering methods require computationally expensive runtime filters that prevent policies from learning safe behaviors autonomously.
Approach
CBF-RL integrates a closed-form control barrier function safety filter with a barrier-inspired reward term during training, allowing the policy to observe corrections and learn to propose inherently safe actions.
Key results
- Proof that continuous-time CBF conditions apply to discrete-time RL via closed-form expressions
- Rapid convergence and robust obstacle avoidance in 2D navigation ablation studies
- Real-world safe stair climbing and obstacle avoidance on a Unitree G1 humanoid robot
- Internalization of safety constraints outperforming filter-only and reward-only baselines
Why it matters
Provides a computationally lightweight, model-free framework for training safe RL policies that can be deployed directly on high-dimensional robotic systems without runtime safety overhead.
Abstract
Reinforcement learning (RL), while powerful and expressive, can often prioritize performance at the expense of safety. Yet safety violations can lead to catastrophic outcomes in real-world deployments. Control Barrier Func- tions (CBFs) offer a principled method to enforce dynamic safety—traditionally deployed online via safety filters. While the result is safe behavior, the fact that the RL policy does not have knowledge of the CBF can lead to conservative behaviors. This paper proposes CBF-RL, a framework for generating safe behaviors with RL by enforcing CBFs in training. CBF-RL has two key attributes: (1) minimally modifying a nominal RL policy to encode safety constraints via a CBF term, and (2) safety filtering of the policy rollouts in training. Theoretically, we prove that continuous-time safety filters can be deployed via closed-form expressions on discrete-time roll-outs. Practically, we demonstrate that CBF-RL internalizes the safety constraints in the learned policy—both enforcing safer actions and biasing towards safer rewards—enabling safe deployment without the need for an online safety filter. We validate our framework through ablation studies on navigation tasks and on the Unitree G1 humanoid robot, where CBF-RL enables safer exploration, faster convergence, and robust performance under uncertainty, enabling the humanoid robot to avoid obstacles and climb stairs safely in real-world settings without a runtime safety filter.