Motion prediction and validation considering perceived risk-based three-dimensional collision avoidance

Predicting human upper extremity motion in three-dimensional (3D) collision avoidance tasks involves integrating biomechanical constraints and cognitive perceived risk into an optimization-based motion prediction framework. The proposed model uses Bayesian Decision Theory to represent human perceived risk stochastically, providing a comprehensive approach to digital human modeling in manual reaching tasks with 3D obstacle collision avoidance. This paper presents an optimization formulation to predict and validate a reaching motion with collision avoidance. First, experimental data was collected in which subjects performed manual reaching tasks involving three distinct 3D obstacles with varying shapes, orientations, and materials. Then, a perceived risk-based 3D collision avoidance model is investigated. Design variables are control points of B-Spline curves representing joint angles for the optimization formulation. The objective function minimizes the joint displacement function and maximizes the end-effector velocity. Constraints include the initial and final postures, joint ranges of motion, upper extremity location related to the experimental setup, and perceived risk-related constraints. The optimization-based framework without perceived risk initially determined the optimal clearance distance, providing a baseline for modeling human motion. This paper modified this baseline through the perceived-risk 3D collision avoidance algorithm to incorporate cognitive factors. Results showed significant improvement in predicting minimum clearance distances when considering perceived risk. For instance, moving around a fragile object caused greater clearance distances, reflecting participants' cautious behavior. The study validated the prediction method by comparing joint angle profiles between experiments and simulations. This work advances digital human modeling by incorporating perceived risk into motion prediction algorithms, moving beyond the traditional reliance on artificial contact spheres. Applications span ergonomics, rehabilitation, and human-robot interaction, offering insights into workspace design, safety, and efficiency. Future research could explore multi-obstacle scenarios, dynamic environments, and alternative loss functions to further refine the model's predictive capabilities.