Imitation Learning and Dimensionality Reduction for the Resolution of Inverse Kinematics in Anthropomorphic Robotic Arms

Arismendi V.A.C.; Aranibar D.B. · Proceedings of the 2024 Latin American Robotics Symposium, LARS 2024 (2024) · DOI: 10.1109/lars64411.2024.10786432

Resumen

This paper presents a novel human imitation-based methodology for solving the inverse kinematics problem in anthropomorphic robotic arms. Unlike traditional approaches, this proposal leverages human intuition and reduces mathematical complexity, facilitating the generation of natural and efficient kinematic solutions. The main innovation lies in a new mathematical approach for dimensionality reduction of human motion data, addressing the challenge of kinematic redundancy. The methodology consists of three phases: dimensionality reduction, nonlinear interpolation, and dimensionality expansion. In the first phase, human motion is captured with an exoskeleton and reduced to a representative variable. Then, a function is created to generalize kinematic knowledge, and finally, the robot's joint angles are obtained. Experimental results, using a 7 degree-of-freedom robotic arm, demonstrate the system's effectiveness in replicating complex human movements in a simulated environment. This research significantly contributes to humanoid robotics by simplifying the calculation process and facilitating the integration of robots into collaborative workspaces.

Instituciones: Universidad Católica San Pablo