Section: New Results

Distributed Estimation and Data fusion

Distributed joint state and input estimation

Participants : A. Kibangou [Contact person] , F. Garin [Contact person] , A. Esna Ashari.

Three consensus-based distributed algorithms have been developed for joint state and input estimation in discrete-time systems. The methods are proper substitutes for distributed Kalman filter in the case in which there are additive faults to the system. Previously developed centralized estimation methods have been reformulated so that the estimator can be used for distributed sensor networks. These new forms are similar to the information form of Kalman filter [34] , [35] . The new forms can be used to propose distributed algorithms based on the consensus of the nodes on calculation of some matrices and vectors. Also a second algorithm is proposed, based on the consensus of the local estimators on local state estimations. This algorithm has less computation effort than the first, but gives a sub-optimal solution in the sense of covariance error. Finally, a third method based on covariance intersection method for diffusing local estimations was proposed in addition. This method also provides a sub-optimal solution. Compared with the second approach, the diffusion of local data is less complicated, however it requires more message communication between nodes.

Data fusion approaches for motion Capture by Inertial and Magnetic Sensors

Participants : H. Fourati [Contact person] , A. Makni.

We are interested to motion capture (or attitude) by fusing Inertial and Magnetic Sensors. In [17] , we present a viable quaternion-based Complementary Observer (CO) which is designed for rigid body attitude estimation. We claim that this approach is an alternative one to overcome the limitations of the Extended Kalman Filter (EKF). The CO processes data from a small inertial/magnetic sensor module containing tri-axial angular rate sensors, accelerometers, and magnetometers, without resorting to GPS data. The proposed algorithm incorporates a motion kinematic model and adopts a two-layer filter architecture. In the latter, the Levenberg Marquardt Algorithm (LMA) pre-processes acceleration and local magnetic field measurements, to produce what will be called the system's output. The system's output together with the angular rate measurements will become measurement signals for the CO. In this way, the overall CO design is greatly simplified. The efficiency of the CO is experimentally investigated through an industrial robot and a commercial IMU during human segment motion exercises. These results are promising for human motion applications, in particular future ambulatory monitoring. The estimated attitude is used to reconstitute the linear acceleration, linear velocity and finally the 3D position from a usual integration procedure (in the case of foot motion) [36] . The problem of attitude estimation is also recently studied within the PhD thesis of Aida Makni. Our goal is to develop a new attitude estimation methods in the case of aerial vehicles (hexa-rotors) by the use of intermittent measures of gyroscopes with the goal to reduce the energy consumption and to gain in the autonomy of the battery.