Tip pose control of low-stiffness tendon-driven cord-like manipulator by machine learning

Abstract

In this study, we develop a manipulator that uses the hose of a vacuum cleaner as its trunk. It can be regarded as a low-stiffness tendon-driven cord-like manipulator. Due to the low stiffness, the restoring force is small compared to the frictional force, resulting in an indefinite pose for a given tendon length. This implies that we need a control that is not based on an elastic model. In this study, we propose a PTP control of the tip by machine learning. The training data were acquired by driving actuators with predetermined pose. Experiments were conducted and evaluated to verify the effectiveness of the proposed method. We verified that the manipulator satisfied the performance requirements for the cleaning task.