OCLC: 809316104
PROQUEST: 2452934059
Potential benefits from human motion understanding range from rehabilitation and physical therapy to ergonomics design, sports training, and computer animation. Robotics-based reconstruction and synthesis of human motion is a powerful tool to study human motion. Understanding human motion requires accurate modeling of the kinematics, dynamics, and control of the human musculoskeletal system to provide the bases for the analysis, characterization, and reconstruction of their movements. These issues have much in common with the problems found in the studies of articulated body systems in robotics research. Task-based methods used in robotics may be leveraged to provide novel musculoskeletal modeling methods and physiologically accurate performance predictions. However, reproducing and synthesizing the basis of human movements in common robotics frameworks bring the following compelling challenges:
In motion analysis, methodologies are developed to characterize human postural behaviors and dynamic skills in a unified framework including task, posture, and additional constraints such as contact with the environment and physiological capacity. Information, which is gained from musculoskeletal models that are mapped into the motion of the human, is exploited in a task-oriented simulation and control framework. Task-driven human performance metrics, including the criteria for operational space acceleration characteristics and human muscular effort, are developed and analyzed for human skills. Using these metrics, optimization criteria are introduced that take into account human skeletal kinematics, muscle activation, physiology, and dynamics, and that correlate to the observed motion characteristics.
In motion control, algorithms are developed to control human musculoskeletal systems in real-time. A marker space control structure is established for the reconstruction of human motion by direct tracking of marker trajectories. Dynamic consistency between marker space tasks, posture, and additional constraints is achieved by recursive projections into the null spaces of higher priority tasks. The human motion control hierarchy is established in marker space following the natural tree-like branching structure of the human musculoskeletal model. The marker space reconstruction methodology allows computing full human motion dynamics in real-time. In addition, an approach for resolving muscle redundancies is developed based on a new hybrid electromyography and conventional computed muscle control method.
These methodologies are validated through three-dimensional dynamic simulations of musculoskeletal models scaled to the subjects. Extensive motion capture experiments are conducted on human subjects of various skill levels including a tai chi master, an elite college-level golfer, a novice golfer, and a professional American footballer for several dynamic movements. Using real-world experimental data, dynamic simulations are exemplarily created and analyzed for golf swings, throwing motions, and gait.
The robotics-based reconstruction and synthesis approaches introduced in this thesis provide an important basis for understanding natural human motion. These tools are applicable to efficient robot control and human performance prediction. Another important application is the synthesis of novel motion patterns in the areas of robotics research, athletics, rehabilitation, physical therapy, and computer animation.