Multi-Objective Reinforcement Learning for Critical Scenario Generation of Autonomous Vehicles