Humanoid's KinetIQ Ascend RL Approach Targets 99.9% Manipulation Reliability

London-based Humanoid has introduced KinetIQ Ascend, a reinforcement learning (RL) approach designed to achieve 99.9% manipulation reliability at human speed and beyond. The method aims to reduce manual tuning from months to days, enabling robots to outperform human demonstrations quickly.

Humanoid’s KinetIQ is a proprietary four-layer AI framework for real-world deployment. KinetIQ Ascend builds on this with trial-and-error learning, allowing robots to improve directly on industrial tasks. According to CTO Jarad Cannon, instead of spending months collecting data and manually tuning skills, the system starts with a basic behavior and refines it into a deployment-ready capability—a process described as a ‘capability factory.’

Founded by Artem Sokolov in 2024, Humanoid has over 250 engineers and researchers, with offices in London, Boston, Vancouver, and San Diego. The company aims to become the No. 1 general-purpose industrial humanoid robotics company within two years. In May, it partnered with Bosch and Schaeffler to scale production of its HMND robots. The article notes that arm drift after long reinforcement training caused by action prefix drift was observed, but no further experimental details or quantitative results are provided.

Source: The Robot Report