A research team from the Massachusetts Institute of Technology (MIT) has developed an open-source platform that allows engineers to study soft robot design. The biologically inspired platform SoftZoo optimizes algorithms for the control and movement of robots. This allows users to automatically generate designs for potential robots.
The platform includes a range of 3D models of animals – including panda bears, caterpillars and fish. They can be used to simulate locomotion, agile turning and path following in different environments. SoftZoo shows the performance differences of the designs on different surfaces such as snow, sand, clay or in water.
“Our framework can help users find the best configuration for a robot’s shape, allowing them to design soft-robotics algorithms that can do many different things,” says MIT graduate student Tsun-Hsuan Wang, a collaborator with the computer Science and Artificial Intelligence Laboratory (CSAIL), which leads the project. “Essentially, it helps us understand the best strategies for how robots interact with their environment.”
Optimize motion control
SoftZoo, in the scientific paper “SoftZoo: A Soft Robot Co-design Benchmark For Locomotion In Diverse Environments” is described and is available on OpenReview.net, is intended to offer a significantly greater range of functions than similar platforms that simulate the design and motion control of soft robots.
The versatility of the framework is based on a differentiable multi-physics engine. It makes it possible to simulate several aspects of a physical system at the same time. The design is optimized by the differentiability of the engine in such a way that the number of simulations, which are often expensive, can be reduced. Users can therefore design and move soft robots with more sophisticated, specified algorithms.
more off-road
The system can simulate interactions with different terrain. In this way, it can be found out which biological structures are better suited to a task than others. This circumvents the problem that robots have difficulty finding their way around in confusing environments because their bodies are not sufficiently adapted to the environment in which they are supposed to operate. SoftZoo now allows the body and its controls to be developed and optimized at the same time. According to the MIT team, this should result in more useful robots in a shorter time.
However, the scientists admit that the transition from simulation to physical robots is difficult. This requires more research, says Wang. The researchers are currently working on improving the platform. SoftZoo should be able to simulate human mechanics, according to the plan. The platform could then be able to test human-like robotic arms that can grip, move and stack objects.
(olb)
