Xiaomi open-sources embodied AI model trained on 100,000 hours of data

Xiaomi's open-source robotics model could accelerate embodied AI adoption across China's manufacturing sector.

Xiaomi's decision to fully open-source its embodied foundation model — trained on 100,000 hours of real-world operation data — threatens to undercut proprietary robotics platforms and accelerate competition in China's fast-growing automation market.

The model, called Xiaomi-Robotics-1, can perform mobile manipulation tasks in unseen environments using natural language commands and requires only small amounts of additional data to adapt to new tasks, the company said in a statement. Xiaomi will release the code and model weights in the near future, a move that could attract developers to its platform.

The model uses cross-embodiment post-training, enabling it to transfer learned skills across different robot hardware platforms — a capability most existing embodied models lack. Xiaomi separately unveiled Xiaomi-Robotics-U0, a 38-billion-parameter multimodal autoregressive foundation model that can generate robot-ready environments from text prompts, adapt existing robot trajectories to new scenes, and produce robot interaction videos from task instructions. By contrast, Thinking Machines Lab's recently released Inkling model, at 975 billion parameters, targets general-purpose AI rather than embodied applications.

For Xiaomi (HKEX: 1810), the move represents a strategic bet on robotics as a growth engine beyond smartphones and electric vehicles. Open-sourcing the model could attract developers to Xiaomi's platform, potentially creating new revenue streams in automation software and services. The global robotics market is projected to reach $74 billion by 2028, with China accounting for more than half of industrial robot installations, according to Goldman Sachs Research.

Open-Source Strategy Targets Developer Adoption

By making Xiaomi-Robotics-1 freely available, the company positions itself against proprietary platforms from Tesla's Optimus and Boston Dynamics, which keep their AI stacks closed. The approach mirrors how Meta's Llama models accelerated open-source large language model adoption — but applied to physical AI rather than text generation. Xiaomi's model, pre-trained on the equivalent of more than a decade of continuous real-world robot operations, could lower the barrier for Chinese manufacturers to deploy automation without building AI systems from scratch.

The open-source release includes both code and model weights, allowing researchers and startups to download, modify, and fine-tune the model for specific applications. This stands in contrast to closed models from competitors that can typically only be accessed through paid APIs or licensing agreements.

Competitive Pressure Intensifies

The release comes as China's robotics sector attracts increasing investment. UBTech Robotics, a Shenzhen-based competitor, has focused on humanoid robots for industrial and service applications, while domestic automakers have begun deploying humanoid robots in EV assembly lines with success rates exceeding 98%, according to industry reports. Nvidia, whose chips power most AI training workloads, stands to benefit regardless of which platform wins, as embodied models require substantial GPU compute for both training and inference.

Xiaomi did not disclose the compute resources used to train Xiaomi-Robotics-1 or specific benchmark results against competing models. The company said the model achieves "open-box usability" for embodied foundation models, meaning it can operate in unfamiliar environments without extensive recalibration.

Investment Impact

Xiaomi shares have gained 23% this year as investors price in the company's expansion beyond hardware margins. The robotics push, if successful, could add $2 billion to $3 billion in annual revenue by 2028, according to estimates from Citi Research. However, the path to monetization remains unclear — open-source models generate indirect value through platform lock-in rather than direct licensing fees. For context, Xiaomi's core smartphone business generated $24 billion in revenue last year, making robotics a small but strategically important bet on future growth.

This article is for informational purposes only and does not constitute investment advice.