Optimal Design of a Humanoid Joint Motor Using a Machine Learning-based Surrogate Model

This paper presents the optimal design of a joint motor customized for humanoid robots. An initial design satisfying the required torque was first established, and a machine learning-based surrogate model was constructed for efficient optimization. The resulting motor design achieved five times the continuous torque of a commercial motor with identical dimensions. Furthermore, demagnetization analysis at high temperatures and currents confirmed stable performance of the optimized motor even under severe environments. Compared to the initial model, the optimal model achieved a 12.22% improvement in maximum torque and a 68.02% reduction in torque ripple at the continuous operating point.