London-based Humanoid Ltd has introduced KinetIQ Ascend, a real-world reinforcement learning (RL) approach the UK humanoid robotics startup says can push its HMND 01 industrial humanoid to 99.9% manipulation reliability at human speed and beyond. The company published a technical report and video results showing bimanual tote handling, machine feeding and picking benchmarks improving within days of training rather than months of manual tuning.
From Demonstrations To A Capability Factory
KinetIQ Ascend sits on top of KinetIQ, Humanoids proprietary four-layer AI stack for real-world deployment. Instead of collecting large volumes of teleoperated demonstrations and hand-tuning each new skill, the robot practices tasks on the line and improves through trial-and-error, refining basic behaviors into deployment-ready capabilities. Chief technology officer Jarad Cannon called the workflow a capability factory that marks how humanoid robots move from impressive demos to tools that industry can actually rely on.
Benchmarks Across Bin Picking And Bimanual Tote Handling
In a machine-feeding test where the robot picked steel bearing rings from a bin and placed them onto a conveyor, KinetIQ Ascend raised throughput by 42%, reportedly running 1.5x faster than the human demonstrations the policy was seeded from. In a cluttered tote hand-off task, throughput rose 85% and success went from 80% to 98%. A bimanual tote lift more than doubled throughput and improved success from 78% to 99%, a roughly twentyfold cut in failures.

Scaling Toward Human-Level Dexterity
Humanoid said KinetIQ Ascend performance improves predictably as training time grows, echoing the compute-scaling behavior of large language models and, together with simulation experiments, suggesting a path all the way to 100% reliability. The team also noted two side findings that matter for factory deployment: improving only the hardest step of a workflow can lift the whole task, and policies generalized to objects not seen during training. Founded in 2024 by Artem Sokolov, Humanoid now has more than 250 engineers across London, Boston, Vancouver and San Diego and has previously partnered with Bosch and Schaeffler to scale HMND production. The announcement lands as rivals like Figure and Neura Robotics push their own reliability, throughput and general-purpose manipulation stories.
Reporting based on coverage from The Robot Report and Humanoid Ltd.
