Risk Factor Quantification for Motion Prediction in a Three-Dimensional Collision Avoidance Manual Reaching Task

Abstract Manual tasks like grasping and aiming require complex coordination between sensory capabilities and motor control. In the presence of obstacles, humans often overestimate the risk associated with collision due to the uncertainty of external stimuli. We present a quantification of the risk factor in multiple manual tasks on the premise that humans can adapt their motion to minimize the loss associated with collision. This study presents data collected for a reaching task around a three-dimensional obstacle chosen from five different obstacles, each with its characteristics. The empirical kinematic data and an optimization-based motion prediction algorithm provide experimental and optimal clearance distances around each obstacle. Based on the statistical decision theory, we define a loss function that will model the human risk perception, and using the collected data, we get an individual loss function depending on human performance around each obstacle. The risk factor is directly related to the clearance distance around an obstacle. It shows that the obstacle’s shape, size, and orientation are defining factors in the trajectory selection by the subjects. The risk factor quantification is the first step towards a comprehensive motion prediction model that could account for obstacle characterization and define trajectories based on what a human would do in similar situations.