InitRech 2015/2016, sujet 1

Summary

This article deals with the benefits of using elastic soft Robots. For now, the approaches to create them are made by inspiration from nature but with their high elasticity it's really hard to control them.
So it's necessary to know how to modelize their movement and that's what the article explain with the Finite Element Method ( FEM ) in order to model the deformations that are the cause of a mechanical coupling.

So, in first the author insist on the difficulty to anticipate the movement because of the infinite number of degrees of freedom, and also on the several advances made on this subject by other scientists.
But to control a soft robot he explains that we need a real-time computation and thanks to algorithms adapted to GPU, the computation time can be reduced.

After that he explains the principle of the FEM. Before everything he needed to know the constitutive law of the soft material by making stress-strain response under tension experiments and he found with a linear regression that the strain/stress ratio is approximatively linear that will reduce the time of online execution of the FEM. But when he assembled the possible rotation in the model is no more linear but it's still possible to compute it with an C++ implementation thanks to the open-source framework SOFA. To have the FEM model we need to compute the following linearization of the internal forces.

But, in addition of the FEM we add a new method that refer new directions in the (J^T) matrix and after that we have to resolve the structure.
The next step is to find λ by using the constraint-based solver that allow us to find the final configuration :
xt = xfree + K^(−1) * J^T * λ.
Where plus definir
So it's possible to obtain actuator model by studying its physical chracteristic to finally knowing the final model.

The next step after knowing the model is to achieve to create the control algorithm. </br>For that we need to know the mechanical coupling between effector and actuator and between actuator to have kinematic equations to build the position control with Gauss Seidel in order to reduce the shift between between actual and desired position. The author also explains that we can add external contact constraints imposed by the nearby object in the Gauss Seidel algortihm to improve the accuracy.

In conclusion, he finishes by showing us two examples of this work : the first one is a mathematics example by simulate some deplacements and check the accuracy with and without collision. The second one is a test on a deformable robot to validate the approach.