Statistical analysis on the human-likeness of 3D reaching movements in humanoid robots

Abstract

Human-like motion is often considered a key feature for intuitive human-robot interactions. In fact, this feature allows human peers to easily predict the robot's intention, which is perfectly aligned with the paradigm of collaborative industries, contributing to more human-centric and resilient industries. The one-sixth power law (1/6-PL) is well known in human motor control. In this work, the Human-like Upper-limb Motion Planner is used to generate three-dimensional (3D) movements of an anthropomorphic robotic arm. By applying direct kinematics, the position and orientation of the hand of the robot is determined. Subsequently, the respective curvature, torsion and velocity are computed. From a total of 600 movements, divided in six sessions, non-linear regression models are fitted and validated, in order to obtain the slope in the log-space of these movements. A statistical analysis of the parameters of the 1/6-PL is performed, and parametric and non-parametric tests are used to compare the results in each of the six sessions.