Research on Motion Retargeting for Humanoid Dual-Arm Robots Using Text-Driven Motion Generation Model

Addressing the challenge of choreographing natural motions for humanoid robots from language, this paper introduces a text-driven motion retargeting system for dual-arm robots. The core innovation lies in a hierarchical optimization strategy that combines motion generation models with prioritized trajectory planning (wrist tracking, elbow similarity, joint norm minimization) under kinematic constraints. Experimental validation on a physical robot demonstrated effective dual-arm motion execution from text commands, achieving semantic accuracy and preserving motion characteristics, though limitations in highly complex bilateral actions due to joint constraints were noted. This work confirms the viability of the hierarchical approach for text-to-motion retargeting and points towards future integration with Large Language Models (LLMs) for enhanced language-driven robot interaction.