Inverse Kinematic Solution Using Neural Networks for Multimodal Inputs and Optimization in Constrained Workspace

Abstract

Kinematics is a fundamental aspect of robotics that deals with the position and orientation of a robotic system. It comprises forward kinematics (FK) and inverse kinematics (IK). While FK is simple, IK is complex due to the nonlinear nature of the equations. Most existing IK solutions are available for unconstrained environments, but few address constrained environments. Constraining the workspace ensures the operation of the robotic system within a predefined safe zone, avoiding potential collisions, which is crucial for applications where safety is a top priority. This letter proposes a method that enables multimodality in a Neural Network model trained to predict the joint angles for a given configuration in a constrained workspace, allowing it to take inputs not just in the form of CSV files of Cartesian coordinates but also images and text inputs. The multimodal nature expands its potential applications across various disciplines where the ability to interpret information in different formats is valuable. This model generates optimal results for all input types and achieves an accuracy of 99% with minimal error.