Step-by-step Realization of Heavy Unstable Object Manipulation by Humanoids based on Motion Failure Detection and Repeated Attempts

Abstract

When a humanoid robot manipulates an object, it is necessary to control the whole-body motion by fully considering the manipulation force corresponding to the object and the reaction force the robot receives during the motion. When the required manipulation force is unknown, there is no choice but to passively control the whole-body. However, when the manipulation force changes rapidly, the robot cannot fully respond and falls down. In order to solve this problem, we have developed a system in which the required force trajectory is acquired based on the previous unsuccessful trial and used in the next trial to achieve a successful motion. As an example of manipulation of an heavy unstable object, we confirmed that the proposed system enables a robot to push up the front wheels a wheelchair by multiple trials, and verified the effectiveness of it.