Abstract
In the Central Nervous System (CNS), the sensory and motor regions exchange information via anatomical connections forming numerous feedback loops which take part in the combination of the sensory signals descending from the various sensory organs. The aim of this study is, in one hand, to extend and generalize the initial models [Droulez and  Darlot 1989, Darlot 1993] describing the sensory information fusion and preparation of the motor orders by the CNS, and on the other hand, to apply these models  to multi - joint arm motion control. Their common structure based on pysiological principles is inspired from the looped connections of the sensori - motor and cerebellar pathways. According to the proposed anatomical interpretation, they would represent mainly the behaviour of the Cerebellum where the coherence and adequacy of neural signals to the real physical quantities are tested. The major  inputs of the model are 3D vectors as follows:

Angular velocity of head in space

Inertial acceleration of head in space

Gaze angular velocity

Linear velocity of head in space

The previous work of Lionel Zupan (Ph.D. thesis presented in 1995) tested the performance of the models by comparing the simulation results of many experimental procedures such as Off Vertical Axis Rotation (OVAR), optokinetic stimulation, barbecue experiments, etc. with the experimental ones. [Zupan et all 1995, Darlot et all 1996]. A first application to the motor control of a body segment (forearm) using the classical equilibrium point principle (Feldman 1986) for representing the loops of intersegmentary regulation was also included in Zupan's work.

The present thesis includes works performed in three areas:
 

Generalization of the proposed principle (SIP) to motor control and combination of the sensory and motor paths:

(Collaboration with  Central University Hospital (CHU) of Caen and  Group of Movement Analysis (GAM) of Dijon)

The initial models were cybernetic circuits consisting of filters, transfer functions, and other well known automatic control elements. We have concentrated our efforts essentially to improve the motor paths which we have extended in two ways:

1. The above mentioned classical elements of the pre motor part were replaced by artificial neural networks simulating  cellular combination of the cerebellar cortex. This work was performed by Nathalie Forestier as a part of her Ph.D. thesis which will be presented in may 1999.

 
2. In this thesis work, the motor part was remodelled in terms of differential equations describing the mechanical and dynamic properties of muscles and joints [J. F. Soechting et all 1997]. The inputs to the motor part were reconstructed so that it works with real muscular activity signals, i.e. the patterns of EMG activity.

Figure 1: Acquisition system (ELITE)

Figure 2: Experiment of point - to - point movement

A series of point - to -point forearm movement experiments in the sagittal and horizontal planes were performed at  the Movement Analysis Laboratory of GAM in Dijon directed by T. Pozzo (Figure 1 and 2) in order to confirm the simulation results  and to determine the anatomical parameters.  The entire model including the combined sensory and motor paths is being put to application to the anthropomorphic robot arm motion control.

Application of the model to the control of  an  anthropomorphic robot arm actuated by artificial muscles (McKibben)

(Collaboration with Group of Automation and Industrial Robotics (GARI) of National Institute of Applied Sciences (INSA) of Toulouse)

The choice of this particular anthropomorphic robot is due to its biological like actuation scheme involving the cylindrical  pneumatic artificial muscles  (McKibben muscles) [B. Tondu and  P. Lopez  1995] , as shown in Figure 3.
 

Figure 3: Experimentation site with an anthropomorphic robot forearm actuated by McKibben Muscles  (GARI , INSA, Toulouse)

In these experiments we have tested the learning capacity of the model including the artificial cerebellar cortex  structure in its premotor path on point-to-point movement. The simultaneous velocity and position control was also achieved.
 

The simulation of motion sickness on the tilting trains:

(Ongoing project by the collaboration of the ENST - ALSTOM - SNCF - LPPA of National Center of Scientific Research (CNRS) - CHU of Caen - Experimental Psychology Laboratory (LPE) of Grenoble)

The sensory integration principle assures the coherence of the estimations (internal to the CNS) of physical quantities exercising on the body and of the kinematics variables of body movements. During the unusual movements, as opposed to usual ones, some sensory organs are stimulated while others are not, and inconsistencies appear between internal signals to the CNS. The model predicts the intensity of these inconsistencies. The experiments on normal subjects showed that the intensity measured by the model is positively correlated with the intensity and the precocity of the motion sickness symptoms, and therefore it allows to predict the occurrence conditions of cinetoses [Denise et coll. 1996].

The aim of this project, performed as a part of this thesis work, is to set up a compact simulator for studying the conditions of motion sickness appearance on electrical high speed (TGV) tilting trains in order to optimize the tilting parameters (such as tilting velocity, tilting angle, etc.). The simulations  were first based on theoretical data calculated from the metric characteristics of the railways namely Limoges - Brive and Lyon - Modane.
 

Figure 4: Measurement helmet

Figure 5: Measuring the head movements in tilting TGV

In order to obtain acceleration and velocity data to be used as inputs to our simulator, we have designed a helmet capable of measuring the linear accelerations and angular velocities of head on 3 axes when worn by a subject during a tilting train ride (Figure 4). The experiments were performed on voluntary subjects on a tilting TGV  prototype during the experimenation campagne (Figure 5) organized by the SNCF and the ALSTOM.