Bridging Mechanical Behavior Differences of Deformable Soft Objects in Simulation and Experiments Using a Data-Driven Model

The complexity of deformable objects poses a challenge when attempting to replicate real-world behavior in simulation, which impedes the use of simulation as a testing environment for empirical applications. This study aims to create a data-driven model for seamlessly translating real-world deformable objects into simulation environments. Compressed soft balls are studied as an example of this strategy. Using machine learning, the model refines simulation parameters based on experimental data, such as forces and contours, allowing for highly realistic simulations and applications in areas such as manipulator manipulation interactions and reinforcement learning for task strategies.