Xiaomi hires ex-Tesla engineer Lu to lead robotic hand R&D

Xiaomi has hired Zach Lu Zeyu, a former Tesla Optimus engineer, to lead its dexterous hand research and development.

Lu, who worked on dexterous grasping and tactile sensing at Tesla, joined the Beijing-based tech firm last month, according to his LinkedIn profile.

He holds a doctorate in mechatronics, robotics, and automation engineering from the National University of Singapore and previously interned at Johns Hopkins University and Tsinghua University.

Xiaomi has recently increased hiring for robotics roles, with a dozen openings for its dexterous hand project and over 200 other robotics-related positions listed online.

The company debuted prototypes of a quadrupedal robot dog in 2021 and a humanoid robot in 2022.

🔗 Source: South China Morning Post

🧠 Food for thought

Implications, context, and why it matters.

Xiaomi’s robotics hiring won’t bring near-term humanoid sales

Xiaomi hired a Tesla Optimus engineer. The move marks a push for humanoid talent yet work stays experimental. CyberOne, shown in 2022, managed greetings and handing a flower, but lagged Atlas and Digit in locomotion 1.
The new Beijing plant can build up to 10 million flagship phones per year and runs 24/7 without people (a lights-out, fully automated line) using 96.8% self-built packaging gear with 100% in-house software 2. That automation is Xiaomi’s near-term robotics business.
Openings for dexterous hand work and 200+ robotics roles support research and development (R&D), not quick product drops. CyberOne mostly acted as a marketing showcase that made Xiaomi’s growing skills clear 1.

Cloud providers can make money from Xiaomi’s open-source embodied AI model

MiMo-Embodied is live on Hugging Face and GitHub 3. Public cloud providers can sell managed fine-tuning or inference hosting. They can also offer deployment for teams building robotics and autonomous driving apps (embodied AI refers to AI that perceives and acts through physical systems like robots, vehicles).
Xiaomi claims the model hits state-of-the-art across 29 benchmarks covering affordance prediction (inferring how an object can be used), task planning, and driving decision-making 3. That track record pulls interest from robotics startups and car makers that want production-grade embodied AI without building a foundation model from scratch.
Simulation platform vendors can bundle MiMo-Embodied with their virtual environments 3. Its cross-domain design enables shared learning between indoor robot navigation and outdoor vehicle perception tasks.