Prioritized Soft Q-Decomposition for Lexicographic Reinforcement Learning