Doctor of Philosophy (PhD)
School of Electrical, Computer and Telecommunications Engineering - Faculty of Informatics
Chen, Yuxin, Transfer of manipulation skills from human to machine through demonstration in a haptic rendered virtual environment, PhD thesis, School of Electrical, Computer and Telecommunications Engineering, University of Wollongong, 2005. http://ro.uow.edu.au/theses/283
Robots are widely used as automation tools to improve productivity in industry. Force sensitive manipulation is a generic requirement for a large number of industrial tasks, especially those associated with assembly. One of the major factors preventing greater use of robots in assembly tasks to date has been the lack of availability of fast and reliable methods of programming robots to carry out such tasks. Hence robots have in practice been unable to economically replicate the complex force and torque sensitive capability of human operators. A new approach is explored to transfer human manipulation skills to a robotics system. The teaching of the human skills to the machine starts by demonstrating those skills in a haptic-rendered virtual environment. The experience is close to real operation as the forces and torques generated during the interaction of the parts are sensed by the operator. A skill acquisition algorithm utilizes the position and contact force/torque data generated in the virtual environment combined with a priori knowledge about the task to generate the skills required to perform such a task. Such skills are translated into actual robotic trajectories for implementation in real time. The peg-in-hole insertion problem is used as a case study. A haptic rendered 3D virtual model of the peg-in-hole insertion process is developed. The haptic or tactile rendering is provided through a haptic device. A multi-layer method is developed to derive and learn the basic manipulation skills from the virtual manipulation carried out by a human operator. The force and torque data generated through virtual manipulation are used for skill acquisition. The skill acquisition algorithm primarily learns the actions which result in a proper change of contact states. Both optimum sequences and normal operation rules are learned and stored in a skill database. If the contact state is not among or near any state in the optimum sequences stored in the skill database, a corrective strategy is applied until a state among or near a state in the optimal space is produced. On-line incremental learning is also used for new cases encountered during physical manipulation. The approach is fully validated through an experimental rig set up for this purpose and the results are reported.