China builds high-precision robot arms for nuclear reactor upkeep

Researchers at China’s Hefei Institutes of Physical Science have developed an AI-driven robotic system capable of performing highly precise maintenance tasks inside fusion reactors, achieving accuracy within 0.1 mm. This system addresses the challenging “peg-in-hole” assembly task critical for replacing reactor components, which traditionally requires slow, manual intervention. Utilizing deep reinforcement learning (DRL), the robot mimics human hand-eye coordination by integrating data from a 2D camera and force/torque sensors, avoiding the unreliable 3D sensors that struggle in the reactor’s high-radiation, reflective environment. To support this AI, the team engineered robust hardware, including a novel robotic joint with an ultra-high reduction ratio (13,806:1) delivering 139 kNm torque, enabling precise manipulation of large reactor components. Additionally, they developed TCIPS, a Transformer-based AI perception model that processes 3D point cloud data to segment the reactor interior into basic geometric shapes, enhancing navigation and obstacle avoidance. These innovations collectively