This paper proposes a data-driven charging and safety protocol design approach using a high-fidelity, physics-based battery model to address the tradeoff between charging speed and battery life degradation. Leveraging the Counterexample-Guided Inductive Synthesis technique, we present a hybrid control strategy that combines reinforcement learning (RL) and data-driven formal methods. We synthesize individual controllers using RL, and then partition the controllers into structures that switch based on initial battery output measurements using data-driven abstraction. The resulting hybrid system combines discrete selection between RL-based controllers with continuous battery dynamics. Once the design satisfies the requirements, abstraction provides probabilistic guarantees on closed-loop performance.
