Bimanual grasping

The current issue is that robots need to be pre-programmed for the specific task they are assigned to. The difficulty in describing a task formally, the high degree of variability and the time-consuming process required to program robots makes this approach less viable in today’s move towards smart manufacturing. One solution is to allow robots to learn by demonstration. Learning by demonstration allows seamless transfer of information from the operator to the robot containing the proven and tested methods of carrying out a specific task. In the project, a direct control teleoperation system was set up to carry out the experiment. The experiment is composed of various sub-tasks such as reaching, moving, assembly, disassembly and rotating of a Tetris block. These sub-tasks are the building blocks of most assembly tasks found in the manufacturing sector. To ensure the accuracy of the experiment, a systematic robot calibration was conducted with the help of a motion-capture system. The object boundary was parametrised to evaluate the point of contact during the various sub-tasks to allow quantification of the grasp physical properties. MATLAB scripts were written to help bimanual grasping analysis and aid task visualisation. Contact between the end effector and object were determined using distance between proxies and force experienced at end effectors; grasp quality was inferred from the proxy velocity and slip analysis.Bachelor of Engineering (Mechanical Engineering