|M.Sc Student||Raz Idan|
|Subject||Comparison of Adaptive Force Control Methods|
|Department||Department of Electrical and Computer Engineering||Supervisor||PROFESSOR EMERITUS Yehoshua Dayan|
|Full Thesis text - in Hebrew|
Industrial robots are required to perform different tasks such as grinding, polishing, buffing, scraping, deburring, twisting and assembly. Robots have an advantage over human workers in performing such tasks, since these tasks often involve stiff materials and surfaces. Handling stiff surfaces is hard labor for human workers, which leads to reduced production, higher costs, a greater chance for accidents and more.
When the robot is in contact with the environment, forces are applied to the environment. For a successful completion of the robot's task, force measurements and force control are required. Since the parameters of the environment are not known exactly, adaptive algorithms are used to compensate for the unknown parameters and to successfully complete the robot's task.
This work addresses the problem of a robot, which is in point contact with a smooth and stiff surface (environment) having parameter uncertainty. The control has to ensure stable tracking of the robot's end effector along a force trajectory normal to the environment, while moving tangentially to the environment in different directions.
A comparison is made among a few different adaptive force control algorithms, recently proposed in the literature. It is assumed that the robot's dynamic model and the location of the environment are precisely known. The robot starts its movement when the end effector is at rest in contact with the environment. The unknown parameter is the stiffness of the surface (environment). In addition, the adaptive methods are also compared to a simple PD controller. The comparison is done by computer simulation.
In the course of comparison, the adaptive algorithms are investigated for tracking different force trajectories. The optimal tuning (values of the control gains) of each algorithm are established by applying cost criteria, subject to weight functions. The adaptive force control algorithms are applied by simulation to three robotic systems: one dimensional robot arm (one degree of freedom), a simple two link planar robot (two degrees of freedom) and the Mitsubishi RV-M2 industrial robot (four degrees of freedom).
The main conclusion which arises from this work is that there is no one superior method for all test cases and all robot models. For every test case, task and robot type one must find the best method to execute the task. No method was superior even in part of the criterion's components. The adaptive algorithms are almost always superior to the simple PD force controller, which performs no adaptation.