Efficient Multibody Interactive Simulation for Soft Robotics (EMISSR)

About:
The EMISSR project aims to create a novel multibody dynamics simulator for soft robotics. The originality of the project consists of a unified constraint-based formulation that can couple rigid and flexible bodies in the same solver. The argument behind a simulator for soft robotics is the fact that traditional robots are not suitable for working in humans and their control has proven daunting especially for nonlinear underactuated systems. The new field of soft robotics is seeing great growth and investment around the globe and Europe needs to seize this economic opportunity. New dynamics simulations and control algorithms are needed for soft robots, that are made up of soft parts and biology-inspired actuators. EMISSR wants to fill in this gap and aims to become a testbed for training control policies based on machine learning and other algorithms. In terms of performance and accuracy, EMISSR will be a high-fidelity interactive simulator, meaning that it will be extremely fast due to its multilevel solver and parallel implementation, while not sacrificing much of the accuracy needed for soft robotics applications. The frictional contact and deformable continua models will not make any unphysical approximations. The elements that will distinguish EMISSR from other existing projects are the two-way coupling of rigid and soft bodies, a fast multilevel iterative solver, stability for high mass ratios and transversal oscillations and robust contact information generation. These will be applied to making virtual replicas of soft robots (e.g. soft body parts, artificial muscles, and tendons, pneumatic tubes), simulating their dynamics and testing machine learning based control policies for large numbers of degrees of freedom.

The EMISSR project aims to develop a new paradigm of high-fidelity multibody dynamics simulation targeted specifically on soft robotics. The need for such a project stems from the robotics industry which is seeing spectacular growth combined with new results from machine learning. Currently, industrial robots are still big and dangerous – their fast-moving rigid parts can easily hurt people around them. Soft robotics can solve this by making robots lighter and out of soft materials. Soft robotics stands for new types of actuation of the soft parts, similar to muscles or other locomotion tissue. Unlike the articulated rigid body dynamics of standard robotics, “current models for soft robots do not capture their dynamics”. Modeling is crucial to developing control algorithms. Moreover, relying on sensory feedback only to control soft robots is not feasible with current technology; so simulation is necessary to train the neural networks and control policies of future robot technology. It is precisely this solution that EMISSR will achieve: a fast, small memory footprint, interactive simulator for multibody dynamics. EMISSR will move beyond existing incipient work by resolving several key issues:

• Fast two-way coupling between rigid and flexible bodies using a unified constrained dynamics formulation,
• Fast iterative solver convergence using a multilevel approach,
• Stability for high mass ratios and oscillations transversal to constraints, and
• Accurate collision detection, producing robust contact information for the solver.