Safe Control Under Input Limits with Neural Control Barrier Functions
This addresses safety in control for complex systems like robotics, but it is incremental as it builds on existing control barrier function methods with neural network enhancements.
The paper tackled the problem of synthesizing safe controllers for high-dimensional nonlinear systems while avoiding input saturation, which can cause safety violations, and achieved a learned control barrier function that maintains safety over nearly 100% of trials on a quadcopter-pendulum system.
We propose new methods to synthesize control barrier function (CBF)-based safe controllers that avoid input saturation, which can cause safety violations. In particular, our method is created for high-dimensional, general nonlinear systems, for which such tools are scarce. We leverage techniques from machine learning, like neural networks and deep learning, to simplify this challenging problem in nonlinear control design. The method consists of a learner-critic architecture, in which the critic gives counterexamples of input saturation and the learner optimizes a neural CBF to eliminate those counterexamples. We provide empirical results on a 10D state, 4D input quadcopter-pendulum system. Our learned CBF avoids input saturation and maintains safety over nearly 100% of trials.