A Neural Network-Based Approach for Trajectory Planning in Robot–Human Handover Tasks

Abstract

Service robots and even industrial robots started recently sharing human workspace for creating new working settings where humans and robots work even hand by hand. On one hand this new scenario raises problems of safety, which are being solved by adding suitable sensor batteries to robot control systems, on the other hand it entails dealing with psychophysical aspects as well. Motion intention understanding and prediction comes more natural and effective if the controlled movement is biologically inspired. In order to generate biologically inspired movements in a robotic assisted surgery scenario, where a robotic assistant shares with, or handover tools to a surgeon, we designed a trajectory planning system based on an artificial neural network (NN) architecture trained on human actions. After the design and training of the neural controller for motion planning, we checked the objective characteristics of the achieved biologically inspired motion as functional minimization (minimum jerk), two-third power law and bell shaped velocity. The controller was also experimentally implemented by using a redundant serial robotic arm (LWR4+, Kuka, Germany), and it was actually perceived as “human-like” in the majority of cases by naïve subjects. The implemented neural-based control strategy provided to be an effective scheme for human-robot interaction control, also by qualitative assessment.