![[photo]](jung.gif)
Systems, Control, and Robotics Laboratory Department of Electrical and Computer Engineering University of California, Davis Davis, CA 95616 (Phone)916-752-3168 (Home)916-758-9775 (Fax)916-752-8428 Located in room 2227 and 2230 in Engineering Unit II. Now at Mechatronics Engineering Department College of Engineering Chungnam National University Taejon, Korea (phone)82-42-821-6876 (Fax)82-42-825-9225 e-mail : jung@meca.chungnam.ac.kr, jungs@hanbat.chungnam.ac.kr
Education
Research Interests
Our main research objective in neural network control area is to use a neural network as a controller or an auxiliary compensator to compensate for any uncertainties in plant. Our control strategy is on-line learning and control. Within this framework, we have developed on-line learning algorithms for training neural controller.
Neural network as a nonlinear adaptive controller is used to compensate for uncertainties in Robot dynamics. Many neural network compensation algorithms have been proposed including feedback error learning(FEL) in the literature. We are proposing a Reference compensation technique(RCT) that neural network compensates at the input trajectory rather than at torque. Learning and control is done on-line. It turns out that the performance of RCT is much better than that of FEL. Besides, RCT has a structural advantage that compensation can be done without modifying internal control structures. Development of a unified neural controller is a future research.
Neural networks are used as compensators that cancel out uncertainties in Robot dynamics and unknown environment(stiffness and position). The unknown environment stiffness can be estimated by sensed force and end-effector position. The new force tracking impedance function that has an infinite compliance is realized. The new impedance function has the capability of tracking desired force by specifying the estimated environment position as a reference. The inaccurate estimation of environment position can also be compensated by a neural network. Therefore, a single neural network can compensate for all the uncertainties occurred when performing impedance force control. The learning signal is developed.
Without using neural network, the robust position control(time -delayed control) is used as an alternative control method. New impedance function that has a capability of desired force tracking can deal with unknown environment stiffness and unknown flat environment surface. If the environment surface is not flat, then a simple adaptive control algorithm can deal with unknown environment surface profile by compensating for uncertainties from inaccurate estimations. Stability and convergence are guaranteed. Experimental results using PUMA 560 showed that the performances were satisfied. This adaptive impedance control is so simple and robust that it is readily used in industrial applications.
The new impedance function can readily be applied to actual industrial applications. Tasks such as deburring and polishing process using industrial robots require sophisticated force control algorithms.
The PIDA controller is an extension of the well known PID controller. It is known that PID controller is not suitable for a third order plant due to lack of one degree of freedom of zero. Adding a zero to PID controller becomes a PIDA(Proportional-integral-derivative-acceleration). The analytical controller design algorithm is developed. Its performance is guaranteed.
Journal Publications
Conference Proceedings
Thesis
Affiliations
Myself
Other Interesting Links
Questions? Comments? E-mail me at
jung@ece.ucdavis.edu