Energy-Regularized Imitation Learning for Force- and Work-Aware Robotic Manipulation

2026-09-16

AI Summary

研究通过训练可微能量预测器来优化机械臂操作，减少物理模拟中的非必要功耗，并在12个任务上验证了方法的有效性。

Abstract

This paper studies energy-aware manipulation as a physically grounded learning problem. We define a joint-space mechanical-work proxy from joint torque and angular displacement, and train a differentiable energy predictor that estimates this work from robot states and actions. The predictor converts a non-differentiable simulator-side physical quantity into a differentiable regularizer for fine-tuning a pretrained manipulation policy. We instantiate the framework with RVT-2 on RLBench and evaluate 12 manipulation tasks involving object contact, articulated motion, placement, pushing, and sweeping. The proposed fine-tuning reduces the average mechanical work from 208.8J to 204.4J (i.e., 2.1% reduction), while the mean task success rate also increases slightly from 86.2% to 86.9%. These results show that work-aware policy optimization can suppress physically inefficient motion without requiring an explicit differentiable dynamics model.

Problem

• Energy-aware manipulation
• Mechanical work
• Policy optimization
• Robotic manipulation

Innovation

• energy-regularized imitation learning
• mechanical-work proxy
• differentiable energy predictor
• work-aware policy optimization