Comparing Human Motion Forecasts in Moving Horizon Trajectory Planning of Collaborative Robots

Abstract

Shared workspaces between humans and robots are a significant part of Industry 4.0. Modern planning algorithms aim to increase human safety in collaborative scenarios and dynamic environments by considering motion forecasts of obstacles. While highly accurate approaches exist, the question arises whether their often greater complexity justifies the benefits that motion planners can draw from them. The paper at hand compares two motion forecasters and evaluates their effectiveness in a robot motion planner to generate collision-free trajectories online. One motion forecaster uses neural networks for predicting, while the other estimates and extrapolates the state of a customized human skeleton. The evaluation shows that simple and online capable motion extrapolations constitute towards human safety in varying situations more than highly accurate neural network predictions.