Periodic Reporting for period 1 - EMISSR (Efficient Multibody Interactive Simulation for Soft Robotics)
Reporting period: 2019-01-14 to 2021-01-13
The problem addressed is that of simulation of soft robots, in particular those made of silicone rubber and actuated by either air pressure or cables. There are three important aspects to this problem: accuracy, speed and flexibility. In terms of accuracy, I emphasized the fact that most traditional simulators cannot handle well incompressible materials like rubber. My method is based on the framework of the mixed finite element method (FEM). It can handle such situations and avoid a dangerous phenomenon called locking (which in brief means simply giving the wrong simulation results). I wrote a simulator in C++ and Python which can simulate the problem in several ways, some potentially faster than many of the existing commercial and open-source solutions. The code was published as open-source and available for anyone to use. It can be further optimized and already allows interactive and even real-time performance for some scenarios. Last, but not least, this simulator (and the theory behind it) allows for a flexible configuration well suited for many of the soft robotics scenarios . For example, one can choose between static and time-dependent simulation, add simple contacts or solver inverse dynamics problems.
I explored ways of unifying constrained dynamics and FEM for elastic simulation. I found mixed FEM as the fitting solution with the added benefit that it fixed locking. I published a preliminary paper on linear and corotational mixed FEM. I investigated pressure actuation scenarios for soft robots and found locking and inversion as two big challenges for getting correct results. I continued developing nonlinear mixed FEM with applications to soft robotics and published the results to a high ranking journal. I worked on the idea that a mixed formulation in terms of principal stretches would work as a generalization for the pressure-displacement mixed formulation and can also prevent locking. In the last part, I focused on using the mixed FEM simulator for inverse problems like estimating the material parameters of rubber or how to actuate the cables in a soft robot to achieve the desired deformation. These latter results have been included in two short papers.
I made available an open-source simulator with non-locking mixed FEM. I found theoretical connections between FEM and constrained dynamics and proposed a new solver that paves the way to unified multibody dynamics. This provided an efficient solution for locking and insights into preventing inversion by using principal stretches and the three-field mixed formulation. All of these will allow soft robotics researchers anywhere to accurately simulate robots made of silicone rubber (actuated via cables or pneumatic means and responding to collisions) at high-fidelity. Also, high performance and fast turnaround times can be achieved by using lower resolution meshes (as well as through code and solver optimizations).