The development of a three-dimensional dynamic posture prediction model for seated operator motion simulation.

A three-dimensional (3D) dynamic model that realistically predicts human postures has been lacking, primarily attributable to the complexity of measuring, analyzing, and modeling 3D multi-segment human movements. This dissertation centers on the development of such a model for seated operator motion simulation by overcoming the various types of complexity. First established was an investigative scheme to alleviate the complexity associated with empirical studies of 3D seated reaching movements and allow experimental identification of significant task effects on dynamic postures. It comprises an analysis method to efficiently characterize movement data, and a design to discern the task factors aside from the intrinsic kinematic variables or constraints. As an illustration, an experiment was conducted to examine two generic task factors: hand motion direction and movement completion time. Results suggest that, within the experimental conditions, the former significantly affects instantaneous postures while the latter does not. The findings clarify the distinction between an instantaneous posture and a static posture, and extend the notion of speed having only a scaling effect to multi-segment movements. A new optimization-based differential inverse kinematics (ODIK) approach was then proposed as a potential means for modeling 3D dynamic seated postures. This approach employs a four-segment seven-degrees-of-freedom biomechanical linkage to represent the human torso and right upper extremity. Kinematic redundancy is resolved in the velocity domain through a weighted pseudoinverse of the Jacobian matrix which maps the joint angle changes to the hand position changes. Sets of weighting parameters quantify an inter-segment allocation strategy that minimizes the total weighted effort, and are estimated by optimization procedures such that the predicted profiles best fit the real motions. This work culminated in the development and validation of a three-dimensional dynamic posture prediction model, using the proposed ODIK approach and a large-scale database of in-vehicle target-directed reaching movements. Systematically validated, the model demonstrates its ability to generate close representations (overall mean angle error $<$ 5 degrees, as averaged over time, multiple angles, and subjects) for the majority of the movements considered, except for the extended reaches towards extreme targets. The model's general success in prediction, computational advantages, and unique characteristics lead to theoretical implications that collectively contribute to a better understanding of the organizational strategies and control mechanisms of human reaching movements.PhDApplied SciencesBiomedical engineeringIndustrial engineeringMechanical engineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/130624/2/9732217.pd