A learning-based inverse kinematics solver for a multi-segment continuum robot in robot-independent mapping

Inverse kinematics (IK) is one of the most fundamental problems in robotics, as it makes use of the kinematics equations to determine the joint configurations necessary to reach a desired end-effector pose. In the field of continuum robot, solving the IK is relatively challenging, owing to kinematic redundancy with infinite number of solutions.In this paper, we present a simplified model to represent a multi-segment continuum robot using virtual rigid links. Based on the model, its IK can be solved using a multilayer perceptron (MLP), a class of feedforward neural network (FNN). The transformation between virtual joint space to task space is described using Denavit-Hartenberg (D-H) convention. Using 20, 000 established training data for supervised learning, the MLP reaches a mean squared error of 0.022 for a dual-segment continuum robot. The trained MLP is then used to find the joints for different end-effector positions, and the results show a mean relative error of 2.90% can be on the robot configuration. Hence, this simplified model and its MLP provide a simple method to evaluate the IK solution of a two-segment continuum robot, which can also be further generalized and implemented in multi-segment cases.