The startup DAIMON Robotics, led by Michael Yu Wang, has launched an omni-modal dataset for physical intelligence. The goal is for robotic hands to sense the world like humans, integrating touch with other sensory signals to achieve dexterous manipulation in real and unstructured environments.
Tactile data for more precise manipulation 🤖
The dataset combines information from cameras, force and contact sensors, along with position and torque data from each joint. This allows robotic models to learn to adapt pressure and grip according to the object. Wang aims to overcome the current limitation of robots, which operate blindly in fine tasks. With this database, systems can anticipate friction and material deformation, approaching human dexterity in manufacturing and assistance.
The day your robot stops crushing the loaf of bread 🍞
Until now, asking a robot to pick up an egg was a culinary Russian roulette. With this dataset, they might stop treating strawberries like bricks. Of course, we'll have to see if they also learn not to confuse a human arm with a test tube. The tactile revolution promises fewer broken objects, although the irony is that the first robot with fine touch will probably use it to avoid making the bed.