Tag
This paper presents a scalable method for end-to-end learning of safe feedback controllers in high-dimensional systems by embedding a control barrier function-based safety filter as an optimization layer, using operator splitting and Jacobian-Free Backpropagation to overcome computational bottlenecks.