A Collision Avoidance Algorithm for Human Motion Prediction Based on Perceived Risk of Collision: Part 2-Application

Occupational Application Digital human models have been widely used for occupational assessments to reduce potential injury risk, such as automotive assembly lines, box lifting, and in the mining industry. Human motion prediction is one of the important capabilities in digital human models, and collision avoidance is involved in human motion prediction. An algorithm proposed earlier was implemented for human motion prediction, and simulated results were found to have a good correlation with the experimental studies. Use of this algorithm can help ensure that human motion is predicted realistically, and thus can impact the accuracy of injury risk assessments. TECHNICAL ABSTRACT Background: With any type of human movement, there is the potential for a collision with other objects. In addition to the objects presented in the environment surrounding one’s body and surrounding the objects to be manipulated, one's own body can become an obstacle. Therefore, consideration of the methods available for avoiding obstacles is necessary to comprehensively describe the way human movements are planned. Purpose: This paper evaluates a collision avoidance algorithm for human motion prediction based on the perceived risk of collision, specifically the application to human motion prediction. Method: Human motion prediction is formulated as an optimization problem with dynamic effort as the cost function, and the perceived risk of collision is considered as one constraint among other constraints. Performance using the new formulation was compared to observed performance from an experiment. Result: Based on the results, the new formulation can account for the suboptimal behavior observed in real subjects while still optimizing biomechanical cost. The predicted motion is much more realistic compared with that from purely biomechanically optimized formulation. Application: The developed collision avoidance algorithm can be applied to optimization-based manual movement prediction in which obstacles need to be navigated.