Enhancing Human Motion Imitation in Humanoid Robots: A Comparative Study of ANN, ANFIS, and GA-Optimized ANFIS for Inverse Kinematics of the Surena-V Humanoid's Arm

Abstract

Humanoid robots are increasingly vital in automation, healthcare, and hazardous environments, where teleoperation enables them to replicate human movements for complex tasks. A key challenge in teleoperation is solving inverse kine-matics, which determines the joint angles required for a robotic arm to reach a specific end-effector position. This research addresses the challenge of solving the inverse kinematics of a redundant robotic arm of the Surena-V humanoid robot, to enhance the robot's capability to accurately imitate human arm movements. Using MediaPipe for human pose estimation, wrist coordinates were input into models to calculate joint angles. Three methods were deployed: Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), and ANFIS optimized using a Genetic Algorithm (GA). The GA-optimized ANFIS model achieved the lowest Root Mean Square Error (RMSE) in joint angle prediction, with values of 6.93, 7.23, 12.33, and 0.9, and superior end-effector accuracy with RMSEs of 0.0199 m, 0.0150 m, and 0.0203 m for the x, y, and z coordinates. These results underscore the effectiveness of GA-optimized ANFIS in handling complex nonlinearities, offering significant improvements in humanoid robot performance for teleoperation and human-robot interaction tasks.