London-based humanoid robot developer Humanoid has introduced KinetIQ Ascend, a training methodology that enables robots to master manufacturing operations through trial and error in real-world conditions. The technology is based on reinforcement learning (RL) and, according to the company, can achieve manipulation success rates of up to 99.9% at speeds comparable to or exceeding human performance.

KinetIQ Ascend expands the capabilities of the KinetIQ framework, which underpins all Humanoid robots. The key difference of the new approach is that RL training is launched not in simulation, but directly on real equipment around the clock. This is the first publicly announced demonstration of end-to-end vision-based RL for VLA (vision-language-action) models trained on a real dual-arm humanoid platform in an industrial deployment setting.

Previously, Humanoid, like most competitors, relied on imitation learning — copying human demonstrations. However, this approach has a fundamental limitation: the model cannot surpass the demonstrator in speed or quality, and it does not learn from its own mistakes. KinetIQ Ascend solves this problem through continuous practice, similar to scaling large language models — the longer the training, the higher the success rate.

Test Results on Manufacturing Tasks

Humanoid conducted A/B testing of KinetIQ Ascend on three real-world operations:

  1. Picking steel bearing rings from a container and placing them on a conveyor. After RL training, throughput increased by 42% — from 291 to 412 rings per hour.
  2. Handing an object from a container to a human. Throughput increased by 85%, average episode duration decreased by 35%, and success rate rose from 80% to 98%.
  3. Dual-arm lifting of a container from a table in arbitrary orientation. After several days of training, throughput increased from 122 to 279 containers per hour, episode time decreased from 22.9 to 12.8 seconds, and success rate rose from 77.6% to 98.9%.

The company also noted that improving the most difficult stage of an operation can boost the overall task result, and the skill transfers to objects the robot did not see during RL training. This is critically important for real manufacturing environments where conditions constantly change.

Industrial Strategy and Competition

Humanoid is actively building a production chain in Europe. In May, the company announced a partnership with Bosch, which will become the contract manufacturer of HMND 01 robots for the European market. Earlier, a phased agreement was signed with Schaeffler to deploy between 1,000 and 2,000 robots at the partner's global sites by 2032.

The announcement of KinetIQ Ascend comes amid an intensifying race in humanoid robotics. Figure has raised over $1 billion at a $39 billion valuation, Apptronik received $935 million from investors including Google and Mercedes-Benz. Since 2024, China has allocated at least $20 billion to robotics development, although actual sales remain limited — around 12,000 humanoid robots, primarily for research purposes.

My expert commentary: KinetIQ Ascend is not just an evolution but a paradigm shift in industrial robot training. The transition from imitation to RL on real equipment solves the fundamental problem of the "ceiling" in speed and quality that is inevitable when copying human actions. If Humanoid manages to scale this approach to thousands of robots in partnership with Bosch and Schaeffler, the company could become a key player in industrial robotics, outpacing competitors that still rely on simulations and demonstrations.