BionicSoftHand Can Rapidly Learn How to Handle Objects with Dexterity

There is a reason robots in science fiction are usually depicted as being humanoid, and it’s not just that it’s easier to put an actor in a robot suit than it is to build a real robot. Everything around us, from the buildings we inhabit to the keyboard you type on, was designed with human physiology in mind. In order for robots to blend seamlessly into our lives, it makes sense to give them a form that lets them navigate the world we designed. Festo’s new BionicSoftHand is engineered to fit our world, and can learn remarkably quickly how to interact with it.

BionicSoftHand incorporates two very important developments: the soft robotics that make up the physical hand, and the artificial intelligence reinforcement learning that allows it to use that hand with dexterity. The hand is constructed from a three-dimensional textile knit made from both elastic and inelastic fibers. By controlling the knit pattern, BionicSoftHand’s creators were able to determine how it would flex as internal chambers are inflated by a special pneumatic valve system.

However, it’s the artificial intelligence system that allows it to utilize that actuation to perform a goal. When a toddler learns to handle an object, it does so by experimenting with it until it’s able to figure out the best method. BionicSoftHand does that with reinforcement learning in a series of digital simulations. Those simulations can be performed over and over again in parallel, so the system can quickly determine the best way to handle an object. Then, when it’s time to manipulate the real, physical object, it can do so with confidence and dexterity.