CRAX: Fast Safe Reinforcement Learning Benchmarking

CRAX (Constrained RL Accelerated with JAX) has been introduced as a fast safety benchmarking framework for reinforcement learning, utilizing the MuJoCo XLA physics engine to achieve significant computational speedups of up to ~100x compared to traditional CPU-based benchmarks. It includes six environment suites and three agent-specific tasks at varying difficulty levels, facilitating large-scale experimentation while highlighting the trade-offs between performance and safety across various safe RL methods. This advancement is crucial for practitioners aiming to efficiently evaluate and prototype safe RL agents in complex real-world applications.