Nightmare Dreamer: Dreaming About Unsafe States And Planning Ahead
This addresses safety concerns for RL applications in robotics, though it appears incremental as it builds on existing model-based and safe RL methods.
The paper tackles the problem of insufficient safety guarantees in reinforcement learning by introducing Nightmare Dreamer, a model-based Safe RL algorithm that uses a learned world model to predict and avoid safety violations, achieving nearly zero safety violations and a 20x improvement in efficiency on Safety Gymnasium tasks.
Reinforcement Learning (RL) has shown remarkable success in real-world applications, particularly in robotics control. However, RL adoption remains limited due to insufficient safety guarantees. We introduce Nightmare Dreamer, a model-based Safe RL algorithm that addresses safety concerns by leveraging a learned world model to predict potential safety violations and plan actions accordingly. Nightmare Dreamer achieves nearly zero safety violations while maximizing rewards. Nightmare Dreamer outperforms model-free baselines on Safety Gymnasium tasks using only image observations, achieving nearly a 20x improvement in efficiency.