Control of Stance and Bipedal Gait in Computer Simulations

Prof. Dr. W. J. Daunicht,

Neurologisches Therapiecentrum

Institut an der Heinrich-Heine-Universitaet

At the NTC Duesseldorf we pursue a project, that aims at the development of a feedback control neuroprosthesis for stance and gait in complete paraplegic patients.

Therefore we compiled a database of bones, joints, muscles, sensors and other human biomechanical data. Further we developed a software package to simulate the movements of complex (bio) mechanical systems with multiple feedback control in 3D. As a tool to implement nonlinear control we developed and implemented DEFAnet, an efficient artificial neural network for high precision function approximation. The main feature of the DEFAnet concept are special architectures combined with special algorithms that allow to determine the synaptic weights of large neural networks in short time. Our control concept is based on the creation of meta-attractors by means of nonlinear control. This allows for larger attraction basins even in the presence of delay.

Our simulations showed that a stable stance of a virtual paraplegic patient (VIP) is possible with independent control of upper and lower body even in the presence of stepping. Also, it has been possible to produce gait of a bipedal walking machine (FREDDY) with a variety of velocities, on uneven ground, and in the presence of a variety of disturbances. We hope that these studies prove useful to develop the next generation of neuroprostheses.

Friday, April 13, 2001

3:30 - 5:00 p.m.

1500 EECS