London-based company Humanoid has unveiled KinetIQ Ascend, an innovative approach to training humanoid robots based on trial and error in real-world production environments. The developer claims this technology will achieve 99.9% successful manipulations at speeds comparable to or exceeding human capabilities.
KinetIQ Ascend expands the existing KinetIQ framework using reinforcement learning (RL). Unlike simple imitation of human actions, the system independently repeats a task, receives feedback on success or failure, and gradually improves its performance. The key difference is running RL not in simulation but directly on real equipment around the clock. Humanoid claims this is the first published demonstration of end-to-end vision-based RL for production VLA models trained on a real dual-arm humanoid platform.
Why Humanoid is betting on RL
Previously, Humanoid, like many competitors, trained systems through imitation of human demonstrations. However, according to the company, this approach has a fundamental limitation: the model cannot surpass the speed or quality of the demonstrator and does not learn the cost of error. KinetIQ Ascend aims to close this gap through practice on real tasks, similar to scaling large language models — the longer the training, the higher the success rate.
Testing KinetIQ Ascend on three production tasks showed impressive results:
- Moving steel bearing rings: throughput increased by 42%, to 412 rings per hour versus 291 for the base model.
- Handing an object to a human: throughput increased by 85%, average episode duration decreased by 35%, and success rate rose from 80% to 98%.
- Lifting a container with two hands: after several days of training, throughput increased from 122 to 279 containers per hour, and success rate from 77.6% to 98.9%.
The company also noted that improving the most complex stage of an operation can boost the overall task result, and the skill transfers to objects the robot did not see during RL training. Importantly, measurements were conducted through parallel A/B comparison with the current base model, which is critical for real production environments where results may vary due to lighting, object positions, or equipment wear.
Humanoid builds an industrial chain in Europe
The company employs over 250 engineers and researchers. In May, Humanoid announced a partnership with Bosch, which will become a contract manufacturing partner for producing the HMND 01 model on the European market. Additionally, the company signed a phased binding agreement with Schaeffler, providing for the deployment of 1,000 to 2,000 robots at the partner's global production sites by 2032.
Competition intensifies
The announcement of KinetIQ Ascend comes amid an accelerating race in humanoid robotics. American company Figure raised over $1 billion at a $39 billion valuation, while Apptronik closed a $935 million round. Since 2024, China has allocated at least $20 billion to robotics development, although actual sales remain limited — about 12,000 humanoid robots last year, mostly for research purposes.
My analysis: Humanoid demonstrates that the key to mass adoption of humanoid robots in industry lies not in simulation but in real-world training on production tasks. KinetIQ Ascend is not just an improvement but a paradigm shift that could give the company a critical advantage in the race for the reliability and speed needed to replace humans on the assembly line.