Universität Bielefeld - Sonderforschungsbereich 360
Recurrent Neural Networks for Real-Time Kinematic Control and
Torque Optimization of Redundant Manipulators
Department of Mechanical and Automation Engineering
The Chinese University of Hong Kong, Shatin, New Territories, Hong Kong
Monday, May 22nd, 2000, 2 p.m., D6 - 135
The use of Kinematically redundant manipulators with more degrees of
freedom than those required for a specific task is a promising approach to
robotic operations. The extra degrees of freedom in redundant manipulators
provide the self-motion which can be utilized to avoid joint limits,
obstacles, singularity, and to optimize various performance criteria. In
this seminar, several recurrent neural networks will be presented for
kinematic control and torque minimization of redundant manipulators. A
two-layer recurrent neural network, called the Lagrangian network, will be
first discussed for pseudoinverse control of kinematically redundant
manipulators through explicitly minimizing the weighted 2-norm of velocity
vector of joints in real time. The Lagrangian network is composed of two
bidirectionally connected layers of neuron arrays representing the
estimated velocity vector and the Lagrangian vector. Secondly, another
two-layer recurrent neural network, called the primal-dual network, will be
discussed for minimum infinity-norm kinematic control of redundant
manipulators in conjunction with a Lagrangian network. Similar to the
Lagrangian network, the primal-dual network also consists of two connected
layers of neurons representing the estimated velocity vector and a dual
layers of neurons representing the estimated velocity vector and a dual
vector. While the command signals of desired velocity vector of the
end-effector are fed into the input layer of a network, the output layer
instantaneously generates the estimated joint velocity vector of the
manipulator with the inverse kinematic equation satisfied. The Lagrangian
network and the primal-dual network will be also discussed for their use in
real-time torque minimization of redundant manipulators based on weighted
2-norm and infinity norm with or without bound constraints on torque. The
analytical and simulation results show that the proposed recurrent neural
networks are asymptotically stable and capable of kinematic control and
torque optimization for redundant manipulators.
Biosketch
Jun Wang is an associate professor and the director of Computational
Intelligence Laboratory in the Department of Mechanical & Automation
Engineering at the Chinese University of Hong Kong. Prior to his current
position, he was an associate professor (1993-1997) and an assistant
professor (1990-1993) at the University of North Dakota, Grand Forks, North
Dakota, USA; a member of technical staff at Zagar, Inc. in Cleveland, Ohio
(1989-1990); a research/teaching assistant at Case Western Reserve
University in Cleveland, Ohio (1986-1989); an instructor of electrical
engineering at Dalian University of Technology, Dalian, China (1986). He
received a B.S. degree in electrical engineering and an M.S. degree in
systems engineering from Dalian Institute of Technology (now Dalian
University of Technology). He received his Ph.D. degree in systems
engineering from Case Western Reserve University. His current research
interests include neural networks and their engineering applications. He
has authored or co-authored about sixty papers published in international
journals, ten chapters in edited books, numerous papers in conference
proceedings. He is an Associate Editor of the IEEE Transactions on Neural
Networks and a senior member of IEEE. Jun Wang is listed in Who's Who in
America, Who's Who in Science and Engineering, Who's Who in the World,
Who's Who among Asian Americans, American Men & Women of Science, among others.
Anke Weinberger, 2000-05-04