Performance enhancement of a 3-DoF humanoid wrist: a dimensional optimization and experimental validation approach

Purpose Wrist mechanisms are crucial for enhancing dexterity and load capacity in humanoid robots, but traditional designs often fail to balance load strength and maneuverability. This study optimizes a previously introduced three-degree-of-freedom (3-DoF) wrist mechanism using a 2PSS-U + RUR parallel configuration. The primary purpose of this study is to conduct multi-objective optimization of geometric parameters to improve overall dexterity and force equilibrium performance, thereby advancing humanoid robot capabilities in real-world scenarios through systematic parameter tuning. Design/methodology/approach The core dimensions of the wrist are analytically modeled in relation to key performance metrics, including dexterity and force equilibrium. Optimization is performed using the Non-dominated Sorting Genetic Algorithm II (NSGA-II), which generates a Pareto optimal solution set for joint angles and linkage ratios. Physical prototypes are fabricated with optimized dimensions, incorporating high-precision actuators and sensors. Experiments evaluate range-of-motion, torque output under controlled loads and dynamic object manipulation tasks to validate the improvements. Findings The optimized prototype achieved peak torques of 8.8 N-m for flexion/extension (F/E), 8.6 N-m for radial/ulnar deviation (R/U) and 1.5 N-m for pronation/supination (P/S), surpassing the non-optimized version. Object manipulation experiments demonstrated enhanced efficiency in complex maneuvers, with reduced energy consumption and improved accuracy in gripping and repositioning objects across orientations, confirming superior dexterity and force equilibrium in practical applications. Originality/value This work pioneers the application of NSGA-II for systematic optimization of parallel wrist mechanisms in humanoid robots, providing a replicable framework that integrates theoretical modeling with empirical validation. It addresses the gap in parameter tuning for such designs, offering enhanced performance metrics and a methodology that can be adapted to other robotic systems, thereby contributing to advancements in robotic dexterity and load handling.