|Ph.D Student||Dejmal Iris|
|Subject||Characterization and Recognition of Manipulative|
|Department||Department of Mechanical Engineering||Supervisor||Professor Miriam Zacksenhouse|
|Full Thesis text|
This thesis investigates the extension of grasp or simple point-to-point hand movements to manipulative hand-movements, which involve the coordination of a large number of joints, and focuses on the characteristics of the intrinsic joint-coordination. Two movement types are studied: simultaneous and sequential, involving forward and backward motions of the same coordination or a series of different coordinated motions, respectively.
For simultaneous hand-movements it is demonstrated that the different joints are tightly coordinated, and the resulting joint-space trajectory is approximately linear. Consequently, this thesis explores the significance of the 1st eigen-vector of the joint-space trajectory in capturing the synergetic structure of the movement. Using the 1st eigen-vector as a characteristic feature, a successful recognition-application is implemented, with an adaptive neural-network, demonstrated for 9 simultaneous hand-movements. The resulting recognition rates are: 97.0, 95.2 and 80.0 for user-dependent (different network for each user), user-independent (single network for all users) and user blind cases, respectively.
The research continues with the investigation of three sequential hand-movements. The quest to divide the complete movement into its basic motions, and to define the nature of each motion, is highly dependent on a segmentation technique. Here, a new segmentation-technique is presented based on the velocity of the initial eigen-functions of the joint-space trajectory. The segmentation reveals two different movement-segments that trace either straight- or curved- joint-space trajectories. The straight joint-space trajectories are generated by task-related motions, and are similar to those generated during simultaneous hand-movements, while the curved trajectories involve repositioning motions.
Two models for hand movements are examined. The first model suggests that the angular motion of each joint is sinusoidal, with a fixed phase-difference between the different joints, and the second model suggests that the angular motion of each joint obeys the Minimum Jerk (MJ) criterion. The two models are compared by simulation to the joints’ path and velocity profile. The paths predicted by the two models are similar but their velocity profiles are different. The MJ model agrees well with the zero velocity constraint at the beginning and at the end of the movement, while the sinusoidal model does not. Therefore, it is concluded that the MJ model is more compatible for describing the hand movements. Both motion-types are demonstrated, straight-line and curved, where the difference in the trajectories shape is attributed to the presence of via points in the repositioning motions, in order to avoid collision with the manipulated object.