Sample-efficient Safe Learning for Online Nonlinear Control with Control Barrier Functions
This work addresses safety-critical control problems in domains like robotics or autonomous systems, offering a novel integration of control theory and learning for improved sample efficiency, though it is incremental in combining existing methods.
The paper tackles the challenge of ensuring safety in online nonlinear control tasks with model uncertainty by proposing a sample-efficient episodic safe learning framework that leverages control barrier functions and optimism-based exploration, achieving provable high-probability safety and near-optimal control performance with formal regret bounds.
Reinforcement Learning (RL) and continuous nonlinear control have been successfully deployed in multiple domains of complicated sequential decision-making tasks. However, given the exploration nature of the learning process and the presence of model uncertainty, it is challenging to apply them to safety-critical control tasks due to the lack of safety guarantee. On the other hand, while combining control-theoretical approaches with learning algorithms has shown promise in safe RL applications, the sample efficiency of safe data collection process for control is not well addressed. In this paper, we propose a \emph{provably} sample efficient episodic safe learning framework for online control tasks that leverages safe exploration and exploitation in an unknown, nonlinear dynamical system. In particular, the framework 1) extends control barrier functions (CBFs) in a stochastic setting to achieve provable high-probability safety under uncertainty during model learning and 2) integrates an optimism-based exploration strategy to efficiently guide the safe exploration process with learned dynamics for \emph{near optimal} control performance. We provide formal analysis on the episodic regret bound against the optimal controller and probabilistic safety with theoretical guarantees. Simulation results are provided to demonstrate the effectiveness and efficiency of the proposed algorithm.