[2M5-OS-19c] 世界モデルと知能(3/4)

[2M5-OS-19c-05] Tactile-Sensitive NewtonianVAE for High-Accuracy Industrial Connector-Socket Insertion

〇Ryo Okumura¹, Nobuki Nishio¹, Tadahiro Taniguchi^2,1 (1. Panasonic Corporation, 2. Ritsumeikan University)

Keywords:State representation learning, Tactile sensor, Connector insertion, Robot control

An industrial connector insertion requires sub-millimeter positioning and compensation of grasp pose of the connector. Accurate estimation of relative pose between a socket and connector is a key factor. World models are promising technology for visuo-motor control. They obtain appropriate state representation to jointly optimize feature extraction and latent dynamics. Recent study shows NewtonianVAE, which is a kind of the world models, acquires latent space which is equivalent to mapping from images to physical coordinate. However, application of NewtonianVAE to high accuracy industrial tasks is open problem. Moreover, there is no general frameworks to compensate goal position in the latent space considering the grasp pose. In this work, we apply NewtonianVAE to connector insertion with grasp pose variation. We adopt a GelSight type tactile sensor and estimate insertion position compensated by the grasp pose. Experimental results show that the proposed method, Tactile-Sensitive NewtonianVAE, outperforms naive combination of regression-based grasp pose estimator and coordinate transformation. Moreover, we demonstrate that domain knowledge induction improves model accuracy.