Developing dextrous hands is central to the current focus on embodied AI for humanoid robots. Human hands represent an extreme level of biological complexity, facilitating simple actions like tying shoelaces or buttoning shirts. Historically, machines have struggled to replicate this advanced human capability. Leading efforts in achieving this breakthrough are currently concentrated within companies in China. Success in replicating the hand is viewed as the final, most difficult hurdle for making robots truly practical products.
A new system named KinetIQ Ascend employs reinforcement learning to achieve human-level dexterity in humanoid robots. This approach demonstrates a high level of reliability, reaching 99.9% manipulation accuracy. It operates effectively at speeds comparable to and exceeding human capability for various industrial tasks. The information was initially published on The Robot Report platform.
League competitions concluded this weekend at RoboCup 2026 in Incheon, South Korea. The event determined the winners across the small, middle, and large humanoid divisions. Top three finishers were recognized for each size class. For example, in the Small division, Invic placed first. Other top teams included Wuhan University from China and Hamburg Bit-Bots from Germany.