Back to News
NewsChinaJobsAi

Humanoid Unveils Robot Training on Real-World Production Tasks

Humanoid has launched KinetIQ Ascend, a new training method for humanoid robots, aiming for a 99.9% success rate in real-world tasks.

Mercation Admin
July 6, 2026 2 min read
Humanoid Unveils Robot Training on Real-World Production Tasks

Humanoid, a London-based robotics firm, has unveiled KinetIQ Ascend, a groundbreaking approach for training humanoid robots through real-world production tasks. The company aims to achieve a remarkable success rate of 99.9% for manipulations performed at or above human speeds.

KinetIQ Ascend builds on the existing KinetIQ framework, employing reinforcement learning (RL) to enable robots to not only replicate human actions but also to independently learn and refine their performance based on feedback from their successes and failures.

Unlike traditional methods that rely on imitating human demonstrations, Humanoid's technology allows for continuous real-time training on actual equipment. This approach is touted as the first documented instance of end-to-end RL based on vision for production models, specifically tested on a dual-arm humanoid platform.

Advancements through Real-World Training

Humanoid's previous training methods limited robots to the speed and quality of human demonstrators, which restricted their potential. The company asserts that achieving higher reliability and speeds necessitates a novel training approach, akin to the scaling of large language models, where prolonged practice correlates with better outcomes.

In trials, KinetIQ Ascend demonstrated significant improvements in three specific production tasks:

  1. The robot's ability to pick steel bearing rings from a container and place them on a conveyor saw throughput rise by 42%, achieving a rate of 412 rings per hour compared to the previous 291.
  2. In a task involving handing an item to a human, throughput increased by 85%, with episode duration decreasing by 35% and success rates improving from 80% to 98%.
  3. When lifting a container using both hands, the robot's throughput improved from 122 to 279 containers per hour, while average episode duration dropped from 22.9 to 12.8 seconds, and success rates surged from 77.6% to 98.9%.

Additional findings indicated that enhancing the most complex steps in operations positively influences overall task performance, and skills acquired during training can transfer to previously unencountered objects.

Humanoid stressed that these improvements were measured through parallel A/B comparisons with the current model, ensuring reliability in real production environments where conditions can vary.

Expanding European Operations

The company, which employs over 250 professionals and has offices in London, Boston, and Vancouver, recently formed a partnership with Bosch to facilitate contract manufacturing for its HMND 01 robots in Europe. This collaboration followed successful tests where robots autonomously managed logistics tasks at Bosch’s facility in Bül.

“This agreement is a critical step in our roadmap, bridging the gap between concept validation and large-scale deployment,” stated Artem Sokolov, founder of Humanoid.

In another significant development, Humanoid announced a phased agreement with Schaeffler to deploy between 1,000 and 2,000 robots at various manufacturing sites by 2032.

Competition in the Humanoid Robotics Sector

The launch of KinetIQ Ascend occurs amid heightened competition in the humanoid robotics market. Notably, American company Figure has raised over $1 billion in funding, while Apptronik secured $520 million in early 2026, adding to its previous funding of $415 million.

China has also invested heavily in robotics, allocating at least $20 billion since 2024. However, actual sales of humanoid robots remain modest, with around 12,000 units sold last year primarily for research purposes.

Recent innovations from Nvidia, Unitree, and Sharpa have introduced new platforms for humanoid robot skill development, while Tether's investment in NEURA Robotics aims to advance mass production capabilities in humanoid robotics.