Problem
Research questions and friction points this paper is trying to address.
safe control
control barrier functions
high-dimensional systems
feedback control
end-to-end learning
End-to-End Learning of Safe Optimal Feedback Control in High Dimensions with Control Barrier Function Layers
🤖 AI Summary
This work addresses the challenge of achieving semi-global feedback control for high-dimensional nonlinear systems under hard safety constraints. The authors propose a novel approach that integrates a control barrier function (CBF)-based safety filter directly into end-to-end policy training. By leveraging operator splitting and Jacobian-free backpropagation (JFB), the method effectively circumvents the computational and differentiability bottlenecks associated with conventional CBF optimization layers in high-dimensional settings. This framework enables, for the first time, safe end-to-end learning in state spaces with over a thousand dimensions. Demonstrated on a multi-agent system with state and control dimensions as high as 1200 and 400, respectively, the approach achieves optimal feedback control that simultaneously guarantees rigorous safety and supports efficient training.
📝 Abstract
We consider the problem of learning high-dimensional semi-global feedback controllers under hard safety constraints enforced by control barrier functions (CBFs). Incorporating CBFs into end-to-end policy training requires embedding a quadratic-program-based safety filter as an optimization layer, but computational and differentiation bottlenecks have largely restricted prior approaches to low-dimensional systems, typically with at most 16 state dimensions. We address this limitation by combining operator splitting with the recently developed Jacobian-Free Backpropagation (JFB) method to enable scalable end-to-end training while preserving hard safety guarantees through the CBF safety filter. We justify this training methodology theoretically using nonsmooth analysis techniques and demonstrate its effectiveness on high-dimensional multi-agent nonlinear control problems with state and control dimensions up to 1200 and 400, respectively.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Control Barrier Functions
End-to-End Learning
High-Dimensional Control
Jacobian-Free Backpropagation
Safe Reinforcement Learning
🔎 Similar Papers
No similar papers found.
