Soft Origami Continuum Robot Capable of Precise Motion Through Machine Learning

Abstract

Soft origami continuum robots show great potential compared with traditional rigid robots because of their hyper-redundant deformation. However, motion control of these robots remains challenging because of their nonlinear kinematics. This letter presents a method based on the multilayer perceptron (MLP) neural network to learn the inverse kinematics of a soft origami continuum robot and make the robot follow the desired motion trajectories. The high compressibility of the origami continuum robot allows the robot to work on different surfaces with thickness variation. The data set comprises 30,240 pairs of valid data (tendon length and tip position). Validation experiments are performed based on static points and typical trajectories (circle, square, eight-shaped curve, lines, and heptagonal spatial curve). Results show that the soft origami robot can achieve precise motion control through the MLP neural network without any sensory feedback. Additionally, the study shows the generalization ability of the developed MLP neural network to move in the workspace outside the data set. The robot has an average position error of approximately 3 mm (1.75% relative to the robot's length) over the workspace.