A hybrid method based on a motion database and motion knowledge for the dynamic prediction of task-oriented human motion

Abstract

Digital human models are more and more frequently employed in product development processes to take human factors into account since the earliest stages of product design. To simulate the interaction of different user populations with a variety of environments, human motion prediction is a useful tool, as it aims at predicting the motion that a generic subject of a user population would reasonably perform to carry out a specific task in a given environment. The motivation of the research work presented in this thesis is the improvement of current motion prediction methods in terms of realism and representativeness. On the one hand, dynamics is included in our formulation, in order to yield physically sound predictions and in view of the fact that the forces and torques acting on and within the human body play a relevant role in discomfort perception. On the other, a hybrid approach is followed, combining the advantages of both data-based methods (which rely on actually performed motions for reference) and knowledge-based methods (which rely on the identification of the motion control laws underlying task-oriented motions). First the method is introduced, and is then applied to the prediction of clutch pedal depression motions. For this purpose, a database of clutch pedal depressions was analysed to gain insight into the subject-related and environment-related features that mostly affect the motion and into the different behavioural patterns that people exhibit carrying out the task. Both a qualitative and quantitative validation of our motion prediction method are presented. The former consists in comparing the most relevant kinematic and dynamic magnitudes in the motion against actually performed motions; the latter is based on the definition of a novel measure, which represents the realism and the representativeness of the predicted motions, and which is compared to the inherent variability of actually performed motions. The results obtained show that the proposed motion prediction method is a valid alternative to current methods, when both the physical soundness and the realism of the motion are required in the prediction.