Safe RL: Comparative evaluation and experimentation