DeepMind and Shadow Robot Indestructible 3-Fingered Robot Hand

Updated: May 10 2024 23:10

Google's DeepMind has partnered with UK-based Shadow Robot to create an incredibly durable and dexterous 3-fingered robotic hand. This innovative piece of technology is set to revolutionize the field of AI research in robotics, paving the way for more advanced trial-and-error learning methods.


The Shadow Robot Company, renowned for its highly advanced 5-fingered Shadow Dexterous Hand series, has now developed a 3-fingered version that is not only precise but also remarkably resilient. The hand can withstand repeated abuse, such as being pounded by pistons from various angles or even being struck with a hammer, making it an ideal tool for the rigorous training processes required in AI robotics.



Weighing in at 9 pounds (4.1 kilograms), with each finger weighing 2.6 pounds (1.2 kilograms), the new hand is designed to closely replicate human dexterity. It can transition from fully open to closed in a mere 500 milliseconds and execute a fingertip pinch with up to 10 newtons of force. This impressive combination of speed, strength, and precision is crucial for the development of AI-powered robots capable of handling complex real-world tasks.


The hand's construction features a blend of durable metals and plastics, including aluminum, brass, acetyl, polycarbonate, and polyurethane flesh, ensuring both strength and flexibility. It operates as a self-contained system, with all actuation and sensing components integrated into the hand and forearm. The Shadow Dexterous Hand employs an electric "Smart Motor" actuation system, incorporating force and position control electronics, motor drive electronics, motors, gearboxes, force sensing, and communication capabilities into compact modules.


To maximize data acquisition for AI training, the hand is equipped with an array of sensors that provide precise data at a rate of up to 1kHz. With 20 motors, each integrated with temperature, voltage, and current sensors, and strain gauges on tendons, the hand ensures meticulous load monitoring. Additionally, 26 Hall effect sensors track joint positions with a resolution of 0.2 degrees, while an IMU in the palm houses a three-axis gyroscope and accelerometer for comprehensive motion tracking.


The development of this indestructible robot hand marks a significant step forward in the field of AI robotics research. By providing a durable and highly capable platform for trial-and-error learning, Google's DeepMind and Shadow Robot are enabling researchers to push the boundaries of what is possible in terms of teaching AI to interact with the physical world.

As Rich Walker, director of Shadow Robot Company, explains, the process of teaching AI to control a robot is similar to that of a baby learning to move and interact with its environment. However, robots are not as resilient as humans, and the early stages of the learning process can be so erratic that robots often shake themselves apart. The new 3-fingered hand's robust design and modular components allow it to withstand the rigors of AI training while still providing the necessary sensitivity for precise object manipulation.



While the hand is currently in manufacturing and no price has been confirmed, Walker states that it will be more affordable than previous models due to its design, which focuses on easy repair and replacement parts. Although developed in collaboration with DeepMind, the hand will also be available to other companies and research institutions. Here is a video showing how researchers are using Shadow Robot's technology for robotics research:


Here is a brief video of Jeff Bezos, founder of Amazon, trying the older version of the robot hand Tactile Telerobot at the last day of Amazon re:MARS Tech Showcase in Las Vegas back in 2019, saying "this is really impressive, the tactile feedback is really tremendous”:


As AI continues to advance and integrate more deeply with robotics, the development of durable, precise, and adaptable tools like the Shadow Robot Company's 3-fingered hand will be essential in pushing the boundaries of what is possible. With this innovative technology, researchers can look forward to a future where AI-powered robots are capable of learning and performing increasingly complex tasks, bringing us one step closer to a world where intelligent machines can work alongside humans in a wide range of applications.