Walker points to the successful collaboration with Google DeepMind, where clear areas of responsibility were defined: Google DeepMind focused on AI, while Shadow Robot developed the hardware. This collaboration allowed both companies to play to their strengths while avoiding the challenges of the other. One of the biggest challenges in robotics is grasping and interacting with objects. Despite advances in artificial intelligence, many systems are still unable to perform complex real-world tasks. Walker explains that her new robotic hand is designed to survive collisions and thus better manipulate objects. This is a significant advance, as traditional robots typically avoid such collisions. The hand is equipped with numerous sensors that provide deep insight into the environment and how the robot is working. This data is used to improve the robot's control. 155 sensors are integrated into each finger, allowing for high precision and flexibility.

Walker explains that the new hand is significantly improved in terms of sensors and mechanics. It has a modular design so that fingers can be replaced without changing the entire system. The hand can typically lift one to two kilograms and exert eight newtons of force at the fingertips. The use of high-quality electric motors from Switzerland contributes to reliability, the developer assures. Interestingly, the hand uses two different tactile sensor technologies: Hall effect sensors and a stereo camera system. The Hall-effect sensors are located on the middle and proximal phalanges of the fingers and measure contact, deformation, and force. The stereo camera system is attached to the fingertips and captures the movement of markers on a soft surface inside the finger. This technology is so sensitive that even the slightest touch is detected. The data collected is used in real time to improve the control of the robots. Walker explains that this is done through machine learning and AI algorithms based on the data provided by the sensors. These algorithms then generate control commands for the robot, enabling precise and adaptive control. The robotic hand is currently being manufactured in Spain, and Walker points out that the hand can be integrated into ROS-1 via a simple API. This makes it easier to adapt and use in different systems. A move to ROS-2 is also planned for the future.

Walker and his team are now looking to scale up and target new industries. The hand is easily adaptable, Walker assures us.