Humanoid debuts KinetIQ Ascend for real world dexterity

Humanoid claims 99.9 percent dexterity at human speed, powered by real world RL.

Testing shows the company’s KinetIQ Ascend reinforcement learning approach is designed to close the gap between lab demonstrations and deployment on real factory floors. Humanoid describes Ascend as an evolution of its KinetIQ platform, built to learn directly from trial and error and convert a basic behavior into a deployment ready capability without the months of manual tuning that once defined industrial robotics projects. The four layer AI framework at the heart of Ascend is meant to help robots reach real world tasks with reliability and speed that rival human operators, while keeping the system adaptable to a range of tasks through ongoing learning. Jarad Cannon, the company’s CTO, framed the shift as a move toward scalable capability rather than bespoke one off solutions. “The humanoid race is becoming a question of scale, and real world RL can be a core part of the answer,” he said. He also noted that the approach can outpace traditional demonstration based methods, with robots improving on industrial tasks within days rather than months.

Humanoid was founded in 2024 by Artem Sokolov and has since grown to more than 250 engineers, researchers, and innovators across offices in London, Boston, Vancouver, and San Diego. The company says its goal is to become the No 1 general purpose industrial humanoid robotics company within two years, a timetable that underscores the push toward production scale rather than lab trials. The HMND family of robots is central to that effort, and Humanoid said it is already pursuing large scale manufacturing partnerships to bring the platform to multiple sites. In May, the company announced a collaboration with Bosch and Schaeffler to scale production of its HMND robots, signaling a shift from prototype validation toward supply chain and factory floor deployment.

From a practitioner’s perspective the development is meaningful for several reasons. First, real world RL that reduces data collection and manual tuning is the kind of capability factory mindset the industry has long pursued. If Ascend can reliably push new skills from an initial baseline into deployment within days rather than months, it changes the economics of robotization for many applications. Second, the move from a lab proof of concept toward a scalable production model hinges on rigorous safety and reliability. Humanoid’s claim that its system can meet human speed on manipulation tasks emphasizes speed but also elevates the need for robust testing and fault handling when things do not go as planned on a live line. Third, the production angle matters. A partnership with Bosch and Schaeffler points to a path where robotics startups must align with traditional manufacturing ecosystems to reach factory scale, including tooling, supply chain, and regional compliance. Fourth, observers should watch how the four layer AI framework translates into continuous improvement in the field. A deployment oriented architecture needs guardrails, clear handoffs between learning and execution, and predictable performance across changing tasks and environments.

In short, Humanoid is turning a bold RL claim into a production strategy. If Ascend can deliver on faster, safer, deployment ready learning at scale, it would shrink the long horizon between dignified demos and everyday factory reliability. The company is betting that a combination of real world learning, a scalable hardware program, and deep partnerships will push it into the lead in a crowded and costly race toward general purpose humanoid automation.