Meta-Learning Control Barrier Functions under Input Constraints

Abstract

This paper studies learning-based synthesis of control barrier functions (CBFs) with an explicit treatment of hard input constraints. Training a task-specific CBF from scratch for each new environment can be computationally expensive and data-inefficient. To address this, we propose a meta-learning approach based on Model-Agnostic Meta-Learning (MAML) that learns an initialization from data collected across a set of related tasks, enabling rapid adaptation to a new environment using only a small amount of task-specific data and a few gradient steps. In closed-loop experiments, a nominal goal-directed control input is computed under input bounds and then filtered by a CBF-QP safety filter to minimally modify the input while enforcing safety under the same actuator limits. Simulation results demonstrate safe, adaptive navigation and good generalization across diverse environments under input constraints.