The second World Humanoid Robot Games officially opened today, with AgiBot's core products Genie G2, Lingxi X2, and Expedition A3 all taking part. In several completed events within the “scenario events” category, AgiBot Genie G2 achieved an impressive record of two gold, three silver, and two bronze medals. As of now, AgiBot sits at the top of the medal table.
This year's Games introduced new “scenario events,” extending competitions into real-world environments across industries including manufacturing, hospitality, households, and logistics. AgiBot Genie G2 won gold medals in both the “Fire Emergency” and “Library” events.
Fire emergency scenario: The competition recreates a real fire emergency. Participating teams must complete three core tasks: identifying hazardous materials, shutting off abnormal valves, and picking up and using a fire extinguisher. The extinguisher weighs 4.5-5 kilograms, far exceeding the load limit of the end-effectors on most humanoid robots available on the market. Instead of choosing an expensive custom gripper, AgiBot added a structural component at the wrist and used the wrist's structural strength to hook up the extinguisher before inserting it into a dedicated carrying bucket on the chassis with a 20° incline, solving the load-bearing challenge.
Library scenario: AgiBot competed as part of a joint team formed with Tsinghua University and Shanghai Jiao Tong University. The event consists of three stages: retrieval and transport, shelving and repositioning books, and identifying and correcting misplaced books. Failure at any stage could interrupt the task, requiring extremely high end-to-end robustness. The team created a high-precision map in advance and calibrated key locations, calculated optimal routes using positioning and navigation algorithms, and used an open-source vision model to identify book spine labels, building a stable and reliable autonomous decision-making system as the robot's “brain.”
According to the official introduction, AgiBot Genie G2 is a mature embodied intelligence product that has been deployed at scale across numerous real-world scenarios. It features force-controlled harmonic-drive joint arms with a rated payload of 5 kilograms, enabling it to handle most objects. Its domain controller delivers 2070 TOPS of computing power, allowing algorithms to run efficiently on the edge and eliminating dependence on external computing resources and networks, resulting in more stable performance.