Xiaomi's new embodied foundation model can execute mobile manipulation tasks in unseen environments using natural language instructions alone.
Xiaomi's Robotics-1 embodied foundation model, pre-trained on 100,000 hours of real-world operation data, enables robots to execute mobile manipulation tasks in previously unseen environments — a milestone that positions the smartphone maker as a full-stack player in embodied AI spanning hardware manufacturing, real-world deployment, and foundation model research.
"This truly achieves a plug-and-play embodied foundation model," the company said in its announcement. The model was further trained with cross-embodiment data to demonstrate stable scalable gains across different robot hardware configurations.
Robotics-1 achieved leading performance across multiple simulation benchmarks and can adapt to new tasks with only a small amount of data, the company said. The model's code and weights will be fully open-sourced in the near future, following Xiaomi's earlier release of Robotics-U0 — a 38-billion-parameter multimodal autoregressive model that unified four embodied AI capabilities including scene generation, trajectory transfer, and robot interaction video generation. That earlier model achieved top scores on the WorldArena benchmark among 126 participating models and improved real-world strategy task completion rates by an average of 26 percent in out-of-distribution conditions.
Xiaomi shares surged 6.3 percent on the announcement, with HK$632.9 million in short selling representing 14.7 percent of turnover. The AI push comes as Morgan Stanley cut its price target 29 percent to HK$32, citing weak EV sales and chip inflation headwinds, creating a tension between the robotics narrative and headwinds in the company's automotive business.
The timing positions Xiaomi in a rapidly intensifying competitive landscape. Embodied AI — systems that perceive, reason about, and act in the physical world — has become a key battleground for companies including Tesla, Nvidia, and Figure AI. Unlike large language models that process text, embodied models must handle real-world physics, sensor noise, and unpredictable environments. Xiaomi's claim of stable gains in unseen environments addresses one of the field's hardest problems: generalization beyond training conditions.
For investors, the robotics narrative provides a growth vector beyond Xiaomi's core smartphone business and its nascent EV operations. The company trades at a premium to traditional hardware peers, reflecting the AI option value embedded in its stock. However, the Morgan Stanley downgrade highlights that EV execution and chip cost pressures remain near-term headwinds that could cap upside, particularly as the company balances R&D spending across three capital-intensive businesses.
This article is for informational purposes only and does not constitute investment advice.