[2O1-GS-8] Robot and real worlds

[2O1-GS-8-04] Learning Compliant Stiffness by Impedance Control-Aware Task Segmentation and Multi-objective Bayesian Optimization with Priors

〇Masashi Okada¹, Mayumi Komatsu¹, Ryo Okumura¹, Tadahiro Taniguchi^2,1 (1. Panasonic Holdings Corp., 2. Ritsumeikan University)

Keywords:Bayesian Optimization, Imitation learning, Segmentation

Instead of traditional position control, impedance control is preferred to ensure the safe operation of industrial robots programmed from demonstrations. However, literature on variable stiffness learning have focused on task performance rather than safety (or compliance). Thus, this paper proposes a novel stiffness learning method that aims to achieve both task performance and compliance. The proposed method simultaneously optimizes the task and compliance objectives (T/C objectives) via multi-objective Bayesian optimization. We define the stiffness search space by segmenting a demonstration into task phases, each with constant responsible stiffness. The segmentation is performed by identifying impedance control-aware switching linear dynamics (IC-SLD) from the demonstration. We also use the stiffness obtained by IC-SLD as priors for efficient optimization. Our experiments on simulated tasks and a real robot show that IC-SLD-based segmentation and the use of priors successfully improve the optimization efficiency compared to the previous baselines.