Soft Robotics Simulations and Soft Bellows Actuators

Abstract

Simulation of soft robots is a challenge in recent decades, due to their in nite degrees of freedom, soft structure and nonlinear behaviours. The majority of available simulation tools are designed for rigid robots that neglect the deformation of materials. This leads to the impossibility of operating simulations of soft robots on these tools. However, some simulation frameworks developed for medical surgery purposes were found to be suitable for simulating soft robots, as all of them are dealing with soft materials. One of the frameworks called Simulation Open Framework Architecture (SOFA) was selected in this thesis to simulate the soft actuators developed by our team. The reason for utilising this framework is due to its open-source structure, reliability of simulation results, specialisation in soft material simulation and the nite element method (FEM) modelling function. To evaluate the performance of SOFA, the stress vs. strain relationship from a soft material was compared with the relationship from the simulated material in SOFA with the same Young’s modulus. The result indicated that the error of the strain of the simulated material was only at 2.9%. The actuators developed by our team, a single bellows actuator and a ring actuator, were later simulated in this thesis. FEM modelling method was utilised to build these actuator simulation models and their performances were tested in a closed-loop control system in several scenarios. A PID control was created and successfully applied in SOFA for testing purposes. The outputs of these scenarios were validated by experimental results and numerical simulation results in the same closed-loop control, and it suggests that a simulation model of a soft actuator in SOFA can have similar behaviours as a physical actuator when they are under the same conditions. For the bellows actuator, both simulated and physical actuators had nearly the same elongation when a positive pressure was applied, and a steady-state error of the simulated actuator was discovered at 1 mm under the PID control, while the steady-state error of the physical actuator was only 0.03 mm. The average steady-state errors in the ring actuator simulation model in a closed-loop control system were distinct according to di erent scenarios, 0.7 mm in the symmetric occlusions and 1.7 mm in the asymmetric occlusions. Additionally, a new collision detection component was developed and integrated with SOFA to reduce overlapping and penetration issues of simulated objects.