PRISM: Personalized Refinement of Imitation Skills for Manipulation via Human Instructions

PRISM is a new AI method that combines imitation learning and reinforcement learning to train robotic manipulation systems using human instructions and feedback. The approach allows generic robotic policies to be refined for specific tasks through natural language descriptions and human corrections, improving performance in pick-and-place tasks while reducing computational requirements.

• →PRISM bridges imitation learning and reinforcement learning for more effective robotic manipulation training.
• →The system uses natural language task descriptions to iteratively generate reward functions for refinement.
• →Human feedback on intermediate rollouts enables policy reusability and improves data efficiency.
• →Testing on pick-and-place tasks showed improved robustness and reduced computational burden compared to methods without human feedback.
• →The approach allows generic robotic policies to be adapted to new goal configurations and constraints.