A post-90s doctoral supervisor from Tsinghua University is working on robots and has raised one billion yuan in a single financing round.

Humanoid robots are at the peak of their popularity.

Star Era, a company affiliated with Tsinghua University, has just secured a financing round of nearly 1 billion yuan. Automobile manufacturers such as Geely and BAIC have joined in, followed by institutions like CDH VGC. For a company founded less than two years ago, this is an uncommon scale of investment.

The key question is: Why Star Era?

Some signals can be seen from the investors. Institutions are betting on technological routes, while automobile manufacturers are betting on manufacturing capabilities. The simultaneous appearance of these two types of capital indicates that investors are already looking for teams that can "mass-produce and implement products in real - world scenarios".

The participation of automobile manufacturers is not accidental. Robots are entering automobile factories to undertake more automated tasks. Automobile manufacturers also hope to occupy a position in the new terminal market. In turn, their supply chains, mass - production experience, and verification scenarios can help robot companies accelerate the implementation of their products.

Currently, industrial capital values two main aspects:

• A technology system integrating hardware and software: Self - developed from large models to the robot body, ensuring strong controllability;

• The ability to implement products across different scenarios: Capable of working in factories, providing services, and running data in scientific research laboratories.

More importantly, its products have begun to be mass - produced one after another.

- 01 -

In August 2023, Chen Jianyu, a 31 - year - old assistant professor (Ph.D. supervisor) at Tsinghua University, left the academic circle to found Star Era. He had been deeply involved in embodied intelligence at Tsinghua for ten years, recruited personally by Academician Yao Qizhi. His research has always focused on "enabling robots to understand the world and act autonomously".

Different from many teams that only focus on models or the robot body, he adheres to the concept of "integrating hardware and software". He often says that a robot is a deeply coupled hardware - software system, and no single breakthrough can make it successful.

Traditional robots have long faced three major difficulties:

They can only perform single tasks and have weak cross - scenario capabilities; they rely on manual programming and lack autonomous understanding; and their hardware is not unified, making it difficult to carry general intelligence.

His judgment is that to solve these problems, "human - like capabilities" are needed. Large models provide the "brain", and a highly flexible robot body provides the "body". Only by combining the two can a truly general robot be created.

This is also the main point of divergence between Star Era and most of its competitors.

In 2024, they launched their first product, the dexterous hand XHand 1. It can operate more than a hundred types of tools and was one of the few domestic actuators that could approach international standards at that time. It was soon used by many domestic and foreign laboratories and humanoid robot companies for data collection.

In 2025, the bipedal humanoid robot L7 was released. The industry generally believes that this marks Star Era's entry into the "first echelon". At the World Humanoid Robot Games, L7 also broke records in events such as high jump and long jump.

More notably, Star Era delivered more than 200 robots in a short period, with more than half of them sold overseas. Given the current pace of the industry, this is an extremely rare progress.

- 02 -

In the past three years, embodied intelligence and humanoid robots have experienced significant technological leaps: Deep learning has brought stronger perception, large models have made generalizability feasible, and hardware costs have also decreased.

The industry is still in its early stage. The core difficulties are concentrated in two areas: the capabilities of the robot body (movement, balance, and fine - grained operation); and data acquisition (insufficient real - world physical interaction data).

These problems make it difficult for traditional special - purpose robots to be used across different scenarios, and they can only play a role in single - point tasks. The industry generally believes that the "humanoid body + end - to - end model" is the main path for the next stage.

Star Era has chosen a path of integrating hardware and software, self - developing the robot body, using an end - to - end model, and creating a data closed - loop. This is the main difference between it and other teams.

Among its competitors, the commonly mentioned leading companies currently include: Zhipu Robot, which focuses on the general humanoid route and deploys multi - scenario products; Unitree Technology, which entered the market through quadruped robots and has strong hardware capabilities; Ubtech Robotics, which has more mature commercialization experience and a complete supply chain system; and Fourier Intelligence, which has long been deeply involved in basic hardware and humanoid platforms.

In terms of market share, although there are several leading enterprises (such as Zhipu, Unitree, and Ubtech), the overall industry is still in its early stage, and the market structure is not yet fully stable.

- 03 -

The underlying technologies of embodied intelligence are evolving rapidly, driving robots from being "able to move" to being "able to understand".

The VLA (Vision - Language - Action) model has become the current mainstream direction. The ERA - 42 model launched by Star Era integrates visual perception, language understanding, and action decision - making in a unified framework, enabling robots to follow instructions, recognize the environment, and perform actions, reducing the errors caused by the fragmentation of previous modules.

The combination of world models and large models is also changing the industry. By introducing environmental prediction capabilities, robots can not only execute tasks but also predict "what will happen next" and make more reasonable plans in complex scenarios. Chen Jianyu believes that this makes robots closer to the rules of the real world, rather than relying solely on data - driven methods.

Hardware costs are also decreasing. As the industry scale expands, key components such as servo motors and direct - drive systems are starting to be mass - produced, and their unit prices are expected to continue to decline. Chen Jianyu believes that there is still room for cost optimization, which will accelerate the large - scale deployment of embodied intelligence from pilot applications.

The focus of competition is shifting from single products to ecosystems. Each company is building developer platforms and opening up SDKs to encourage third - party participation in function expansion. The company that can form a more complete ecosystem is more likely to establish industry standards.

Driven by these trends, embodied intelligence is entering more scenarios: flexible manufacturing, retail services, elderly care assistance, scientific research and education, inspection and security, etc. Its role is also gradually shifting from an "automation tool" to a "collaborative intelligent agent".

This article does not constitute any investment advice.

This article is from the WeChat official account "Pencil News" (ID: pencilnews), author: Truth - teller. Republished by 36Kr with permission.