The main advances during this period concerned the experimental part of the project and more specifically the calibration part. In the first part of the project, all the tests have been performed using the commercial Adept controller of the Quattro robot. In order to implement the new complete kinematic models that will be used in simulation, the development controller has to be opened.  

A common session between the different partners was then held at Lirmm facilities in order to make the first tests with the models that have been translated from Matlab to C. After preliminary validation tests, a first series of measurements was performed in the new development environment. 

The Dual Space Computed Torque Control aims at controlling the Adept Quattro parallel robot in the Cartesian space, in order to improve the accuracy of the motions.

Without the platform and the outer arms, the Cartesian error in the direction of the motion is divided by about a 4.4 factor. In Z-direction, the error is divided about by 5. However, we have noticed that the orientation error is multiplied by a factor from 1.25 to 1.71, which should be reduced in future work. In the case of combined motions in the X- and Y- directions together, instabilities appear and will be investigated. An experiment using the Dual Space Control with the outer arms and the platform let hope to correct well inertial effects of the platform thanks to Cartesian feed forwards.

Those preliminary results of the Dual Space Control are really promising, we hope improving a lot the performances of the Adept Quattro robot.

Development of different control schemes for the redundantly actuated Quattro S650 parallel manipulator.

The proposed controllers were implemented and integrated of the simulator of the robot. Extensive simulations were conducted for different situations to show the performance of each controller. This step will be very useful for the next one concerning the experimentation of the proposed controllers in real-time on the industrial robot.

With the kinematic calibration, the future issues will be the implementation of the new models in the Adept controller and perform the measurements in order to validate the improvement brought by the calibration (65% in the mean error and 50% in the maximum error within the workspace).

Development of different (kinematic and dynamic) control schemes for the Quattro S650 parallel manipulator. These controllers were implemented in the simulator of the robot. Extensive simulations were conducted for different scenarios and simulation conditions.       

With the kinematic calibration, we improve the precision of our robot of 65%, considering 33 parameters to describe the Quattro robot. The study of the sensibility of each parameter allows us to suppress some of them without interfering with the results quality. With 25 parameters we obtain 60% of improvement and with 21, 56%.

We wish to investigate the influence of the sampling of the calculation of the Quattro Jacobian on an adaptative control law. For the same trajectory simulation and the same set of Cartesian gains, we studied the influence of various Jacobian computation frequencies. For a frequency higher than 200 Hz, the system remains stable but below 200 Hz, the system does not response correctly anymore. We suspect the low frequency of the Jacobian computation to be one cause of the instability of the control.

The first step was an improvement of the duration of the computation of the Jacobian matrix. The second step was to to interpolate the Jacobian coefficients.

The Jacobian is now competed at the highest frequency as allowed with the current CPU load. The micro-interpolation of the Jacobian at 8 kHz has finally allowed solving some instability problems.

Cartography of the robot workspace was done and experimental measurements have been performed. Using these measurements, and after the validation of the optimization algorithm in simulation, three models were calibrated and compared in terms of their accuracy improvement and sensitivity of parameters. A more complete model using 31 parameters led to an improvement of 70% in the static error value.