Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 28, 2023 - July 01, 2023
To support and replace all kinds of tasks that people perform on a daily basis, multi-degree-of-freedom humanoid robots that can perform multiple tasks with a single robot are expected to become a reality. However, since the development cost of motion teaching for humanoid robots is enormous, an intuitive motion teaching method is required. One example of an intuitive motion teaching method is teleoperation using a bilateral remote device, but at present it is only used for motion teaching to robots with simple mechanisms. In this study, we teach a multi-degree-of-freedom humanoid robot using a bilateral remote device, and verify whether towel capturing motion can be generated using deep predictive learning. Through experiments on actual equipment, we confirmed it is possible to properly take in a towel placed at an unteached position. This shows the effectiveness of complex motion teaching and deep predictive learning using a bilateral remote device.
