A Collision Avoidance Algorithm for Human Motion Prediction Based on Perceived Risk of Collision: Part 1-Model Development

Abstract

OCCUPATIONAL APPLICATIONS Digital human models have been widely used in occupational biomechanics assessments to prevent potential injury risks, such as automotive assembly lines, box lifting, patient repositioning, and the mining industry. Motion prediction is one of the important capabilities in digital human models, and collision avoidance is involved in human motion prediction. We propose an algorithm that will ensure human motions are predicted realistically, and finally, use of this algorithm could help enhance the accuracy of injury risk assessments using digital human models. TECHNICAL ABSTRACT Background: Humans perform daily tasks such as reaching around an obstacle with ease, even though the complexities of such behavior are largely hidden from those performing them. Optimization-based motion prediction has employed numerical methods in order to predict human movements. However, these movements are heavily constrained, such that the planning of the motion is explicitly provided in the optimization formulation of the problem. This implies that for each task a unique optimization formulation is needed, which relies heavily on the experience of the code developer to provide these constraints. Purpose: Cognitive psychology has focused on the reasoning or motivation behind the planning of movements and provides an opportunity for digital human modeling to adopt these theories to provide a more general or versatile motion prediction framework. Humans tend to overestimate the risk associated with colliding with objects during movement. We present the formulation of a collision avoidance algorithm that considers the perceived risk, for future use in a human motion prediction application. Methods: An experiment was completed to evaluate human performance when avoiding obstacles during movement. Using Bayesian inference, perceived risk was modeled and minimized for use in human motion prediction. Results: The experimental results were used to derive a formula in which the perceived risk associated with the task could be quantified in a gain/loss context. Overestimation of risk by a subject was modeled using the observed behavior and the results of simulations based on the parameterized risk model are presented. Conclusions: The algorithm presented, based on the perceived risk of collision, can be integrated into human motion prediction to generate realistic human motion considering collision avoidance.