London-based company Humanoid has introduced KinetIQ Ascend, a revolutionary approach to training humanoid robots based on trial and error directly on real-world production tasks. The developer claims this technology can boost manipulation accuracy to 99.9% at speeds comparable to or exceeding human performance.
KinetIQ Ascend extends the existing KinetIQ framework. Unlike traditional imitation learning, where a robot simply copies human actions, the new method uses reinforcement learning (RL). The system independently repeats a task, receives a success or error signal, and continuously optimizes its behavior based on this feedback. The key difference is that RL runs not in simulation but around the clock on real equipment and in production scenarios. Humanoid states this is the first published demonstration of end-to-end vision-based RL for VLA models trained on a real dual-arm humanoid platform in deployment conditions.
Why Betting on RL is a Paradigm Shift
Previously, Humanoid, like many competitors, relied on imitating human demonstrations. However, as the company rightly notes, "a model that copies demonstrations cannot exceed the speed or quality of the demonstrator and does not learn the cost of error." The final percentages of reliability and the transition to superhuman speed require a different approach. KinetIQ Ascend bridges this gap through continuous practice on real tasks. This resembles the scaling of large language models: the longer the training, the higher the success rate.
Testing KinetIQ Ascend on three production tasks yielded impressive results:
- Bearing ring stacking: After RL training, 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%.
- Two-handed container lifting: After several days of training, throughput grew from 122 to 279 containers per hour, average episode duration dropped from 22.9 to 12.8 seconds, and success rate climbed from 77.6% to 98.9%.
The company also noted that improving the most complex stage of an operation can enhance the overall task result, and the skill transfers to objects the robot did not see during RL training. The improvement was measured through parallel A/B comparison with the current base model, which is critical for real production environments where results can vary due to lighting, object positions, or equipment wear.
Industrial Expansion and the Race of Giants
Humanoid is actively building an industrial chain in Europe. The company employs over 250 engineers and researchers. Recently, a partnership with Bosch was announced, which will become a contract manufacturing partner for producing HMND 01 robots for the European market. Additionally, a phased binding agreement was signed with Schaeffler, providing for the deployment of 1,000 to 2,000 robots at the partner's global production sites by 2032.
The announcement of KinetIQ Ascend comes amid an accelerating race in humanoid robotics. Figure raised over $1 billion at a $39 billion valuation, Apptronik closed a $935 million round, and China has allocated at least $20 billion to the industry since 2024. However, actual sales remain limited — about 12,000 humanoid robots were sold last year, mostly for research purposes.
Expert opinion: KinetIQ Ascend is precisely the step that separates laboratory experiments from industrial deployment. The transition from imitation to self-learning from real mistakes is the only path to creating truly autonomous robots capable of working in unpredictable environments. If Humanoid succeeds in scaling this approach, we will witness the beginning of a new era in manufacturing automation.