Shadow Robot recently finished a multi-year project working with Google DeepMind to produce a new class of robot hand – reliable and robust enough to survive the challenges of reinforcement learning in the real world, while at the same time dexterous and capable enough to perform human-like manipulation tasks.
The new robot hand has a number of new technologies that can be used in the development of next generation robot hardware, from high speed control architectures based on force-controlled N+1 actuation, to new stereo tactile fingertips capable of sensing the lightest of touches and surviving the high forces typical of grasping and manipulation research.
Shadow Robot tested the new robot hand for thousands of hours – from component-level wear tests to high force self-collision tests and impact testing – and produced a modular robot finger capable of assembly into a wide range of robots with serviceability and reliability at the core of the design.
In this session, we will discuss the design challenges, how we mapped the development space, the technical and practical solutions we developed to address the challenges, and show how these solutions could be applied to other robotics problems.