Humanoid's KinetIQ Ascend: Redefining Robotic Dexterity

London-based robotics company Humanoid unveiled KinetIQ Ascend, a reinforcement learning framework that achieves 99.9% manipulation reliability at human speed, boosting throughput by up to 85% in industrial tasks. The system allows robots to learn from trial and error within days, potentially transforming industrial automation.

Humanoid's KinetIQ Ascend: Redefining Robotic Dexterity Humanoid's KinetIQ Ascend uses reinforcement learning to boost robotic capability, achieving human-level manipulation speed and reliability. In the often understated field of robotic manipulation, a London-based company named Humanoid is quietly making waves. Last week, they introduced KinetIQ Ascend, a reinforcement learning /glossary/reinforcement-learning approach that aims to achieve 99.9% manipulation reliability at human speed and potentially beyond. Scaling the Humanoid Race Jarad Cannon, CTO of Humanoid, claims the humanoid race is now a matter of scale, and real-world reinforcement learning could be the solution. "Robots that once demanded months of manual tuning are outperforming human demonstrations within just days," he says. It's a bold claim. Could this be the key to moving humanoid robots from impressive demos to reliable industrial tools? Founded by Artem Sokolov in 2024, Humanoid is racing to become the leading general-purpose industrial humanoid robotics /category/robotics company within two years. With over 250 engineers and researchers from top global tech firms, the company is expanding its influence, with offices in London, Boston, Vancouver, and San Diego. The 'Capability Factory' KinetIQ Ascend is described as a 'capability factory' by Cannon. This four-layer AI framework allows robots to learn through trial and error, refining basic behaviors into deployment-ready skills. The real-world applications are vast, from single-arm picking tasks to complex bimanual handling. In practical tests, Humanoid's robots have shown impressive improvements. For instance, in a machine-feeding task, the throughput increased by 42%, and robots operated at 1.5 times the speed of initial human demonstrations. In another task, picking items from a cluttered tote, throughput jumped by 85%, with success rates soaring from 80% to 98%. A New Era for Robotics? The implications of KinetIQ Ascend are significant. Robots aren't only improving within days but are also generalizing to objects they haven't seen during training /glossary/training . This brings up a critical question: Are we on the brink of a new era where robots can finally offer the dexterity and reliability industries have long sought? Humanoid's approach suggests that enhancing the most challenging part of a workflow can elevate the entire task. The idea that the company’s method might scale to 100% reliability is nothing short of fascinating. But, does this mean robots will soon replace human hands in more industries? While the container doesn't care about your consensus mechanism, the prospect of robots reaching such efficiency levels presents a thrilling yet challenging future. The ROI isn't just in the model. It's in the dramatic reduction of time and error rates in everyday tasks. Get AI news in your inbox Daily digest of what matters in AI.