Fusion IK: Solving inverse kinematics using a hybridized deep learning and evolutionary approach

Abstract

Inverse kinematics is a core aspect of robot manipulation. This paper presents an approach to solving Inverse Kinematics (IK) for robots, including articulated industrial ones, combining deep learning with an evolutionary algorithm. Fusion IK passes the manipulator’s target and current joint values into a neural network, the results of which are then used to seed an evolutionary algorithm, Bio IK, to complete the solution of the inverse kinematics problem. Fusion IK allows for solving the position and orientation of the robot while attempting to minimize joint movement times. Comparisons between Fusion IK and its underlying algorithm Bio IK are tested on a six-degree-of-freedom articulated industrial robot as well as a 20-degree-of-freedom robot to explore the move times that Fusion IK produces. The comparisons show that the variations of the Fusion IK algorithm show comparable results to its underlying evolutionary Bio IK algorithm on a six-degrees-of-freedom articulated robot and improvements on a 20-degree-of-freedom robot without any additional hyperparameter tuning. The results show that Fusion IK could be of real value regarding the movement time and the quality of the obtained solutions upon further research, especially with higher degree-of-freedom robots.