人間ロボット協調作業のためのベイズ最適化によるインピーダンスプロファイル生成

抄録

Impedance control is a widely used control method for physical human-robot collaboration systems. This paper proposes a new method to generate the damping profile to improve the performance of human-robot collaboration. First, monotone cubic interpolation is used to make smooth damping trajectories from several data points. The damping values at each data point are then optimized using bayesian optimization. Bayesian optimization is modified to consider task failures, and discretization of parameter space is used to improve the optimization process. Finally, experiments are conducted with a 7-DOF robot arm in a simplified 2-dimensional manipulation task, and results showed that the proposed method help improve the task performance.