BIPED LOCOMOTION PROJECT

Fig. 1: Tripp, a passive walking robot

Tripp, is a simple 4 DOF robot, and provides a platform for abstractly studying the issue of dynamics, separate from control. However, in human locomotion the two issues are intrinsically intertwined. To study both the issues concurrently, a 10 DOF, actuated, 3D biped robot is currently being developed. It will provide a realistic physical model of human legs, and be used as a testbed for studying the interaction of dynamics and neural control.

Fig 2: Joint Control using NNetview. The network is part of the NSCM model used to vary the angle
and stiffness of a joint.The green connections indicate positive weights, blue connections negative.

Understanding of the physical dynamics provides the foundation for studying the neural mechanism of locomotion, which is the central goal of this project. A Neural Spinal Cord Model (NSCM), which embodies state-of-the-art knowledge of spinal neural architectures involved in locomotion, is currently being implemented using artificial neural networks, in NNetview software developed by Neuronics . This is being done in cooperation with physiologists and experts in physiotherapy, at the Paraplegic Center of the University Hospital, who provide the empirical support for the model.

The current model allows coordinated movement of two hip and knee joints. The central neural circuit consists of an oscillator which adjusts the rhythm of the movement according to the feedback from the muscle spindles, a neural joint control network consisting of motoneurons 1a- and 1b-Interneurons, simulated muscle spindles and Golgi tendon organs. The 1a-related reflex controls the lenght of the muscle, the 1b-reflex the muscle tension.

Meanwhile an interface between the neural simulation and the Lokomat of the ParaCare Lab of the Univervity Hospital has been built. This machine helps paraplegic patients to exercise movements. Our neural network is able to control the Lokomat robot. So far we have successfully controlled one joint of the locomat with the neural model.

Fig 3: Graph of Neural activities. The graph shows simulated activities of flexor and extensor
motoneurons together with the angle of the hip and the phase of the gait movement.

We are now extending the model by adding adaptive components to the network. This extension will require only few additional neurons. The learning ability will be based on the usage of the feedback of the movement (e.g. contact of the foot with the floor) which allows to serve as a measurement of "success" of the movement and at the same time to build up an implicit neural "model" of the optimal movement under certain conditions.

Once it is developed, the neural model will be implemented as a controller for the 10 DOF biped robot. Its performance on the physical platform will allow us to test, evaluate and refine the model via an iterative process including comparison between analytical experiments and synthetic results. The understanding of the human neural system gained in this project, will be used by the Paraplegic center to improve locomotor therapy for partially paraplegic patients, and for Functional Electrical Stimulation of walking patterns in completely paralyzed patients.

Fig 4: The Lokomat driven gait orthosis, used
for rehabilitation of paralyzed patients.