PIAP advanced methods for sensor fusion over the last year. Since soft manipulators have no rigid links, the modelling of their shape is not trivial. Soft manipulators can bend at any point, twist and elongate as well. External forces can act at any point of the arm and its deformation is distributed not only at discrete points. The above causes the previously used shape calculating algorithms to be insufficient and, because of that, new methods had to be developed.
For the purpose of shape estimation, a new data fusion algorithm has been developed by PIAP.
The structure of the data fusion system is presented in the figure above. The system takes as input the data from various sensors such as length sensors, force sensors, pressure sensors and vision. The sensor data is then analyzed along a set of steps. The first step involves the chamber length values to be used for a rough approximation of the manipulator configuration. Since the direct force measurement is not available, module shape approximation is required to convert the experienced moments into forces. Knowing the estimated force values, more complex shape calculation method can be employed. The deformation of the manipulator consists of variable bending, torsion and elongation. Because these methods of shape approximation use the estimated force value, the result can be still not very precise. To correct this, we use vision. Replacing the whole shape reconstruction system by a vision system is not possible since there is a high probability that some parts of the manipulator will be occluded. Nevertheless, the position of the visible parts is well-defined. The improvement is performed by adjusting the force approximation using an inverse kinematics algorithm until the simulated shape matches the visible manipulators parts.