Grasping and Master Gloves

This project is generally concerned with both autonomous control and teleoperation of the Utah/MIT Dextrous Hand. Current hardware includes a right hand, a 6-DOF Cartesian robot table, and a Utah Dextrous Hand Master.

Fingertip Force Control

Actual demonstrations of robot grasping are far behind theories of grasp control, and what is needed is to implement even the simplest schemes well rather than develop yet more unimplementable theories. The approach has been to characterize the hand's drive system, in order to understand mechanical limitations in finger control and to improve performance insofar as possible by model-based low-level control.

The actuators, comprised of electropneumatic servovalves and pistons, were modeled and identified experimentally, and a model-based controller was developed (Henri and Hollerach, 1994).
The dynamics of the tendon drive system was experimentally characterized (Nahvi and Hollerbach, 1994).
Feedback and feedforward control schemes were developed and contrasted for actuator, tendon, and fingertip force control (Nahvi, M.Eng. thesis, Dec. 1994).

Based on these studies, it has been determined that tactile sensors are required for accurate fingertip force control. We plan to incorporate capacitance-based tactile sensors developed by Sarcos for this hand. Finally, we are hoping to obtain a left hand to study two-handed manipulation, since the utility of using just one hand is limiting.

Task-Oriented Sensor-Based Manipulation

This project concerns the development of advanced sensing systems combined with sensor based processing, planning and control strategies for autonomous robotic manipulation, inspection and assembly tasks. One aspect is to develop an instrumented gripper for the Sarcos Dextrous Arm, that incorporates proximity sensing and tactile sensing. The tactile sensors are being developed in conjunction with Sarcos, which has developed capacitance-based tactile sensors for the Utah/MIT Dextrous Hand (Johnston, Zhang, Hollerbach, and Jacobsen, "A full tactile sensing suite for dextrous robot hands and use in contact force control," Proc. IEEE Intl. Conf. Robotics and Automation, Minneapolis, April 22-28, 1996, in press.)

Teleoperation with Master Gloves

The Utah Dextrous Hand Master, a prototype exoskeleton-based position master, is used to measure finger joint angles. This hand master is the most accurate one available, because of the use of four-bar linkages. The hand master has been used to control the Utah/MIT Dextrous Hand, while the Bird magnetic sensor (Ascension Technology) controls the position and orientation of the Cartesian robot holding the hand. In order to achieve a high level of performance in this teleoperation, three problems were identified and addressed.

Kinematic mismatch between the human hand and robot hand. A Cartesian fingertip mapping algorithm was developed to circumvent this mismatch (Rohling, Hollerbach, and Jacobsen, 1993). Key features over past approaches include handling workspace boundaries explicitly, mapping orientation as well as position, and a three-level priority scheme if the desired human fingertip position and orientation is not exactly realizable.
Inaccurate kinematic model of the human hand. An open-loop calibration method was applied to obtain an accurate kinematic model of a human operator's hand (Rohling and Hollerbach, 1993, 1994). This approach is much more accurate than past schemes which relied on caliphers and bony landmarks. The 1994 IEEE International Conference on Robotics and Automation version of the paper received the Anton Phillips Best Student Paper Award.
Control errors by the robot hand in following the commanded trajectory. This aspect has been addressed in the fingertip control studies above.

The robot hand is presently attached to a three degree-of-freedom Cartesian robot table, with a 3-roll wrist based on Rosheim's simple hollow wrist. The two parts of this robot were products of different design projects involving students at McGill University. In the longer run, we wish to develop force reflection and tactile stimulation in a hand master.