Dual-Objective Reinforcement Learning with Novel Hamilton-Jacobi-Bellman Formulations