UC San Diego

Control Barrier Functions (CBFs) have gained rapid popularity in the recent years as a method to verify and enforce safety properties in safety-critical controllers for autonomous systems. However, developing a valid CBF that is not overly conservative can prove to be a non-trivial task in conjunction with input constraints. Using a recently developed algorithm called RefineCBF, this task can be made easier by providing a constructive method that iteratively constructs a valid CBF using dynamic programming (DP) based reachability analysis. This work seeks to validate that RefineCBF can be used with hardware-in-the-loop by demonstrating the algorithm successfully enforcing safety online for a robotic agent. We successfully demonstrate this by showing that a three degree of freedom robot can safely reach a goal pose in the presence of obstacle with minimal violations to safety using a safety filter whose constraint is informed from RefineCBF. Additionally, we demonstrate that in scenarios where the obstacles change in time in a non-adversarial way, RefineCBF can be used to adaptively enlarge the safe set online.