LEAP-CBF: A Safety Filter for Uncertain Systems with Least-Effort Adversarial Potentials
The paper proposes a safety filter that measures how much disturbance effort it would take to push a robot system into failure.
LEAP is presented as a control barrier function for undisturbed systems and as a way to handle bounded cumulative disturbances without the usual robust-CBF conservatism. The authors build LEAPs with on-policy deep reinforcement learning. They test the method in simulated multi-agent systems, then validate it on quadruped and quadrotor hardware. ArXiv · AI/CL/LG's note
LEAP is presented as a control barrier function for undisturbed systems and as a way to handle bounded cumulative disturbances without the usual robust-CBF conservatism. The authors build LEAPs with on-policy deep reinforcement learning. They test the method in simulated multi-agent systems, then validate it on quadruped and quadrotor hardware. ArXiv · AI/CL/LG's note
score 4