MORE | Spring 2022

Modeling Deviation in Bending Direction of a Soft Arm Manipulator Using Regressive Neural Networks

The compliant nature of a soft robot’s material and design offers certain challenges when it comes to modeling and control. The lack of rigid structures allows the system to bend along its entire length and in any direction instead of at specific joints, like in a rigid robot. Moreover, the actuation, generally pressure controlled, does not directly correlate with the bending observed in the robot. These two factors result in a non-linear deviation in the bending direction of the robot which makes it difficult to set up a controller for it. This study focuses on modeling said error by regressively solving for it using neural networks.