AI-driven robotics

17:05 - 17:10

Abstract

Physical artificial intelligence refers to robotic systems that can interact in a human-like fashion with the physical world. Physical AI aims to perform work in the real physical world, like doing your laundry or handling delicate objects. These systems must sense their environment with smart sensors to inform their real-time decisions. Robots can already see and hear thanks to decades of engineering development in audio and vision processing. However, their sense of touch remains rudimentary compared to humans. The emerging field of tactile sensing calls for an interdisciplinary collaboration between sensor and robotic engineers. This pitch presents the joint work between Melexis, a supplier of integrated sensors, and IMEC/Brubotics. Melexis developed an integrated magnetic force sensor, measuring the 3D contact force between an object and a robot, achieving a 10 mN resolution (equivalent to 1 g). IMEC/Brubotics integrated an array of such sensors into a gripper to enable fragile-object manipulation. The collaboration covered the sensing chain end to end: the mechanical contact force, the raw magnetic signals, the subsequent processing algorithm, and integration into a physical AI system. The complete demonstrator features the three sensing modalities coordinated by generative AI: audio for speech recognition, vision for motion planning, and tactile sensing for real-time feedback. This work demonstrates how compact 3D magnetic force sensors can enhance the physical capabilities of robots.