London-based Humanoid unveiled KinetIQ Ascend on July 5, 2026, a real-world reinforcement learning method targeting 99.9% manipulation reliability at human speed or faster. It sits on top of the company's four-layer KinetIQ AI framework.
The reported results, all self-published: on a machine-feeding benchmark moving steel bearing rings from bin to conveyor, throughput rose 42% and the robot ran at 1.5 times the speed of the human demonstrations it learned from. On cluttered-tote picking with hand-off to a person, throughput rose 85% and success went from 80% to 98%. On bimanual tote lifting, throughput more than doubled and success went from 78% to 99% — roughly a twentyfold cut in failures, after days of training.
The humanoid race is becoming a question of scale, and real-world RL can be a core part of the answer. Robots that once required months of manual tuning are now outperforming human demonstrations within days. — Jarad Cannon, CTO, Humanoid
The claim underneath the numbers
Cannon describes the method as starting from a basic behaviour and letting reinforcement learning refine it into a deployment-ready capability — "a process we describe as building a capability factory." The company reports that performance scales predictably with training time, following an LLM-like curve, and that improving only the hardest sub-step lifts whole-task performance. Robots also generalised to objects they had not seen.
Humanoid was founded by Artem Sokolov in 2024 and has more than 250 engineers and researchers across London, Boston, Vancouver and San Diego. It partnered with Bosch and Schaeffler in May 2026 to scale production of its HMND platform, and closed a $152 million Series A in July 2026 at a $1.35 billion valuation.
Key Facts
- KinetIQ Ascend unveiled July 5, 2026 — real-world RL targeting 99.9% manipulation reliability at human speed or faster
- Machine feeding: throughput +42%, running at 1.5x the speed of the human demonstrations
- Cluttered-tote pick and hand-off: throughput +85%, success 80% → 98%
- Bimanual tote lift: throughput more than doubled, success 78% → 99% — about a 20x reduction in failures
- Humanoid: founded 2024 by Artem Sokolov, 250+ engineers across London, Boston, Vancouver and San Diego; Bosch and Schaeffler partners since May 2026
Why it matters
Every humanoid pitch eventually runs into the same wall: demonstrations are cheap and reliability is not. Ninety percent success is a video; 99.9% is a job. Humanoid's contribution, if the results hold outside its own benchmarks, is evidence that the last stretch can be closed by training time rather than by hand-tuning — which is the difference between a research project and a manufacturing process.
The caveat is that all of this is self-reported on internally chosen tasks. Independent replication on shared benchmarks is what would turn a capability-factory claim into a fact.
Frequently Asked
What is the headline result?
Bimanual tote lifting went from 78% to 99% success after days of training — roughly a 20x cut in failures.
Is this independently verified?
No. The benchmarks and figures are the company's own.
Why does it matter?
It suggests the gap between a demo and a deployable task can be closed with training time rather than manual tuning.