UK-based startup Humanoid has developed the Kinetiq vision-language-action AI system that teaches humanoid robots new physical skills in days [1].

This advancement could accelerate the deployment of automation in sectors like automotive plants and warehouses. By reducing the time required to program complex movements, the technology allows robots to integrate into existing human-centric workspaces more efficiently [1].

Dr. Alex Patel, CEO of Humanoid, said the Kinetiq model reduces training time from months to days [1]. The system is designed to enable robots to perform repetitive tasks in factories and warehouses that were originally designed for human workers [1].

Research into the system's efficiency highlights a specific method called "touch-dreaming." Dr. Maya Liu, a lead researcher, said the touch-dreaming approach boosted success on five tricky tasks by 90.9% [2]. This capability allows robots to refine their physical interactions with objects more rapidly than previous iterations of robotic learning.

The development comes as the broader humanoid robot race intensifies. Other companies are also reporting milestones in autonomous operation. John Doe, CEO of Figure AI, said Figure AI’s Helix-02 can now run full eight-hour autonomous shifts in a factory [3].

While some reports suggest training times for certain skills may be reduced to hours [2], the company's primary claim remains that the process now takes days rather than months [1]. This shift represents a significant departure from traditional robotic programming, which often required exhaustive manual coding, or lengthy trial-and-error periods in simulated environments.

“"Our Kinetiq model reduces training time from months to days."”

The transition from month-long training cycles to day-long cycles suggests a shift toward 'general purpose' robotics. If AI can rapidly adapt a robot's physical behavior to new tasks without extensive reprogramming, these machines become viable for dynamic industrial environments where tasks change frequently, rather than being limited to a single, static assembly line function.