Arm drift after long reinforcement training caused by action prefix drift. Source: Humanoid
Robotic manipulation is making progress with artificial intelligence. London-based Humanoid last week introduced KinetIQ Ascend, its reinforcement learning, or RL, approach designed to reach 99.9% manipulation reliability at human speed and beyond.
“The humanoid race is becoming a question of scale, and real-world RL can be a core part of the answer,” stated Jarad Cannon, chief technology officer at Humanoid. “Robots that once required months of manual tuning are now outperforming human demonstrations within days.”
Humanoid is building humanoid robots with the goal of becoming the No. 1 general-purpose industrial humanoid robotics company within two years. Founded by Artem Sokolov in 2024, it has more than 250 engineers, researchers, and innovators from top global tech companies.
With offices in London, Boston, Vancouver, and San Diego, Humanoid said it is building commercially viable, scalable, and safe systems for real-world applications. In May, the company partnered with Bosch and Schaeffler to scale production of its HMND robots.
KinetIQ Ascend supports ‘capability factory’
Humanoid said KinetIQ is its proprietary four-layer AI framework designed for real-world deployment. KinetIQ Ascend builds on the previous KinetIQ platform with trial-and-error learning, helping the company’s robots improve directly on industrial tasks.
“KinetIQ Ascend, our real-world RL method, offers a new approach to developing robot capabilities,” said Cannon. “Instead of spending months collecting data and manually tuning every new skill, we can start with a basic behavior and allow RL to refine it into a deployment-ready capability – a process we describe as building a ‘capability factory,’ which marks how humanoid robots move from impressive demos to tools that industry can actually rely on.”
Submit your session idea for the 2026 RoboBusiness
Humanoid tests demonstrate improved manipulation
Humanoid tested KinetIQ Ascend on several tasks, including picking parts from bins, handing objects to humans, and lifting and moving containers using two robot arms. It has proven effective across a range of manipulation scenarios, claimed the company.
In a machine-feeding application, a robot picked steel bearing rings from a bin and placed them onto a conveyor. KinetIQ Ascend reportedly increased throughput by 42%, enabling the robot to operate at 1.5× the speed of the human demonstrations it originally learned from.
A different task involved picking items from a cluttered tote and handing them to a person. The same approach increased throughput by 85% while improving success rates from 80% to 98%. Across increasingly complex manipulation scenarios, KinetIQ Ascend continued to deliver significant improvements, said Humanoid.
In a third bimanual tote-handling task the robot lifted a tote from a table using both arms. KinetIQ Ascend more than doubled throughput, and success rates rose from 78% to 99%. This represented a roughly twentyfold reduction in failures, with all results achieved after only a few days of training.
Humanoid said the results demonstrated that KinetIQ Ascend shows a new way of developing robot capabilities, proving effective across a range of real-world operational tasks, from high-speed single-arm picking to complex bimanual handling.
KinetIQ Ascend also demonstrated that robot performance improves predictably as training time increases. It’s similar to how large language models (LLMs) improve as more compute and data become available. The company said that the observed scaling trend, supported by simulation experiments, suggests that its method scales all the way to 100% reliability.
A new approach also revealed two additional findings: improving only the hardest part of a workflow can improve the entire task, and robots were able to generalize to objects they had not seen during training.
Humanoid outlined all these findings in a new technical reportwhich provided the full methodology behind KinetIQ Ascend, including the training infrastructure, algorithmic solutions, and a deeper analysis of the results.
GIPHY App Key not set. Please check settings