STAIR: Addressing Stage Misalignment through Temporal-Aligned Preference Reinforcement Learning

However, its effectiveness in multi-stage tasks, where agents sequentially perform sub-tasks (e.g., navigation, grasping), is limited by stage misalignment