Humanoid Robotics Leap to 98 Percent Success Rate in Assembly

From Manuel Christa Manuel Christa | Translated by AI 2 min Reading Time

Related Vendors

Xiaomi reports significant progress in humanoid robots at its own car factory. Within a few months, the success rate at an assembly station increased from 90.2 to 98 percent. At the same time, the company introduces two AI models for embodied intelligence.

98 percent correct: According to the company, the rate is now one percentage point below the human level.(Image:  Xiaomi)
98 percent correct: According to the company, the rate is now one percentage point below the human level.
(Image: Xiaomi)

As recently as March, Xiaomi had reported 90.2 percent for a three-hour autonomous run of the humanoid at a cycle time of 76 seconds. According to the company, the jump to 98 percent is only one percentage point below the human level. However, it remains unclear how many cycles were included in the evaluation, what exactly counted as an error, and whether rework or human intervention was taken into account. An independent review of the metric is not available.

Advances Also in Flexible Workpieces

Beyond the assembly task, Xiaomi reports results for workpieces that are more difficult to grasp and place. According to the company, the robots achieve around 90 percent when sorting side panels of the center console and returning parts containers. Xiaomi describes these runs as the first long-term tests with flexible workpieces in a car factory.

Malleable or irregular parts are considered challenging in robotics because they change their position and shape, making them harder to calculate than rigid components. Xiaomi regards the fact that the machines achieve stable rates over longer periods in this area as evidence that the systems are transitioning from the lab to operational use.

Two AI Models for Embodied Intelligence

On the software side, Xiaomi released two separate models. The U0 model with 38 billion parameters achieved an overall score of 73.64 on the WorldArena platform and ranked first among 126 tested models. It is designed to help robots perceive their environment and plan actions.

The second model, Xiaomi-Robotics-1, was trained on 100,000 hours of operational data from more than 1,700 scenarios and achieved a success rate of 57.4 percent on the Benchmark RoboCasa365.

Both publications demonstrate how strongly Xiaomi relies on feeding real operational data from its own production back into the development of its robots. This data foundation from ongoing manufacturing is considered essential for making the systems increasingly reliable step by step. 

Subscribe to the newsletter now

Don't Miss out on Our Best Content

By clicking on „Subscribe to Newsletter“ I agree to the processing and use of my data according to the consent form (please expand for details) and accept the Terms of Use. For more information, please see our Privacy Policy. The consent declaration relates, among other things, to the sending of editorial newsletters by email and to data matching for marketing purposes with selected advertising partners (e.g., LinkedIn, Google, Meta)

Unfold for details of your consent