The Woodruff School

SUMMARY

This thesis proposes the development of a teleoperated bilateral system with supervisory control capabilities. The bi-manual robot system’s morphology is optimized to improve its work-space visibility to the operator, manipulability, and kinematic performance. Further, an optimization framework to automate routine interaction tasks that involve coupled force and motion is presented.
The manipulability, the kinematic and dynamic performance of a robot system are dependent on the robot morphology. We use expert kinematic and kinetic data collected through an experimental setup to optimize the robot system’s manipulation capability for a defined set of meso-macroscale
manipulation tasks. The optimized morphological parameters are used in motion planning to improve the kinematic performance (i.e., design informed motion generation). The modeling, optimization methodology, hardware implementation, and preliminary analysis of this method will be presented. Moreover, to automate routine tasks for supervisory control, a hierarchical control method is proposed to generate the control input sequence while satisfying task, state, and control
constraints to ensure safety (e.g., interaction with soft bodies). The proposed method uses a distributed cost by using the Alternating Direction Multiplier Method. In the implementation, the optimal control sequence is generated in a Model Predictive Control fashion to accommodate realtime
execution. This high-level trajectory optimizer is followed by a low-level force admittance controller to adapt to any uncertainties that arise from motion disturbances and variations in the contact material properties. The performance will be demonstrated in a physics-based simulation environment and hardware in real-time with simulated disturbances. The results of the proposed morphology optimization method will exhibit that the global robot dexterity, work-space visibility to the operator, and kinematic performance of the system are improved
relative to a benchmark robot design and the results will be demonstrated in real-hardware.
Further, demonstrations of automated tasks will be expected to exhibit satisfactory task completion while satisfying state, control, and task constraints in real-time execution.

SUBJECT:	Ph.D. Proposal Presentation

BY:	Lasitha Wijayarathne

TIME:	Friday, September 11, 2020, 10:00 a.m.

PLACE:	BlueJeans Meeting ID: 461 364 990, bjn.vc

TITLE:	Design Informed Motion Optimization, and Shared, Supervisory Control of a Dexterous Bilateral SystemDesign Informed Motion Optimization, and Shared, Supervisory Control of a Dexterous Bilateral System

COMMITTEE:	Dr. Frank L Hammond III, Chair (ME) Dr. Ye Zhao (ME) Dr. Jaydev Desai (BME) Dr. Charlie Kemp (BME) Dr. Jeffrey Bingham (Google X)

SUMMARY This thesis proposes the development of a teleoperated bilateral system with supervisory control capabilities. The bi-manual robot system’s morphology is optimized to improve its work-space visibility to the operator, manipulability, and kinematic performance. Further, an optimization framework to automate routine interaction tasks that involve coupled force and motion is presented. The manipulability, the kinematic and dynamic performance of a robot system are dependent on the robot morphology. We use expert kinematic and kinetic data collected through an experimental setup to optimize the robot system’s manipulation capability for a defined set of meso-macroscale manipulation tasks. The optimized morphological parameters are used in motion planning to improve the kinematic performance (i.e., design informed motion generation). The modeling, optimization methodology, hardware implementation, and preliminary analysis of this method will be presented. Moreover, to automate routine tasks for supervisory control, a hierarchical control method is proposed to generate the control input sequence while satisfying task, state, and control constraints to ensure safety (e.g., interaction with soft bodies). The proposed method uses a distributed cost by using the Alternating Direction Multiplier Method. In the implementation, the optimal control sequence is generated in a Model Predictive Control fashion to accommodate realtime execution. This high-level trajectory optimizer is followed by a low-level force admittance controller to adapt to any uncertainties that arise from motion disturbances and variations in the contact material properties. The performance will be demonstrated in a physics-based simulation environment and hardware in real-time with simulated disturbances. The results of the proposed morphology optimization method will exhibit that the global robot dexterity, work-space visibility to the operator, and kinematic performance of the system are improved relative to a benchmark robot design and the results will be demonstrated in real-hardware. Further, demonstrations of automated tasks will be expected to exhibit satisfactory task completion while satisfying state, control, and task constraints in real-time execution.