"Star Motion Era" has completed a nearly 300 million yuan Pre-A round of financing, focusing on the native robot large model + AI hardware platform | Exclusive by HardcoreTech.
Author | Nan Huang
Editor | Silai Yuan
HardcoreTech exclusively learned that the embodied intelligence company "Star Motion Era" has recently completed a nearly 300 million yuan Pre-A round of financing. This round is jointly led by Qingliu Capital, Yuanjing Capital, and Alibaba, with Ceyuan Capital as the follower, and the existing shareholders Lenovo Ventures, Century Golden Resources, Jinding Capital, Zeyu Capital, Qingkong Tiancheng continue to invest. China Renaissance continues to serve as the exclusive financial advisor for this round.
The financing funds will be used to accelerate the original and breakthrough research and development of embodied intelligence technology and the commercialization process of general-purpose humanoid robots, and to continuously build the company's talent and technology barriers.
This is also the third round of financing completed by "Star Motion Era" since its establishment; previously, the company received an angel round of financing of over 100 million yuan led by Lenovo Ventures, with Jinding Capital, Zeyu Capital, and Qingkong Tiancheng as followers, and the existing shareholder Century Golden Resources as the additional investor, as well as a seed round of financing of several tens of millions of yuan led by Century Golden Resources and followed by Turing Ventures.
STAR 1 running on the Danxia landform
"Star Motion Era" was founded by Jianyu Chen, an assistant professor at the Institute for Interdisciplinary Information Sciences of Tsinghua University. It is the only humanoid robot enterprise in which Tsinghua University holds shares and is supported by the Shanghai QiZhi Institute. The company focuses on the research and development of the native general-purpose embodied intelligence agent and has launched a new hardware platform defined by the native robot large model + AI.
The native robot large model is an AI-native learning paradigm proposed by "Star Motion Era", which is both versatile and simple, and has sufficient Scale-Up (expansion) capabilities. Its algorithm is based on full-modal input fusion to process sensory information such as vision, hearing, and touch, which can effectively restore the interaction mode between the organism and the physical world, enhance the ability to process cross-modal information, and enable the intelligent agent to perceive and understand the environment more comprehensively and flexibly.
STAR 1 playing with a cat
STAR 1 performing a grasping movement on the desktop objects
According to the actual measurement performance, in terms of versatility, the robot can conduct self-exploration and self-learning through the native robot large model, truly achieving autonomous evolution and wide application in multiple scenarios. In terms of simplicity, the model adopts an end-to-end architecture, enabling the intelligent agent to adapt to tasks and environments in real-time without the need for complex programming or human intervention, significantly improving flexibility and development efficiency. Currently, this model has been applied in the sixth-generation humanoid robot STAR 1 of "Star Motion Era".
The movement and operation capabilities of Xingdong STAR 1 adopt an end-to-end pure learning method. With only one neural network, the robot can generalize walking and grasping in different environments. Based on its language-vision-action large model, actions are directly generated through language and visual signals, which can effectively avoid information transmission loss and support more efficient Scale-Up to support the robot in new task scenarios to master new skills with a small amount of data, improving the adaptive ability and scalability.
End-to-end AI algorithm, long-time domain control
In terms of hardware, "Star Motion Era" has launched a hardware architecture for humanoid robots that integrates software and hardware. From the bottom-level motor to the bipedal system and the dexterous hand, all components are based on the premise of being AI-friendly, ensuring that the algorithm can be quickly verified on the hardware, significantly improving the Sim2Real (from simulation to reality) efficiency, and achieving the co-evolution of software and hardware. For example, in STAR 1, the degree of freedom has been increased from 20 to 55, which can support more complex humanoid postures; its joint torque has also been increased from 150N·m to 400N·m, and the joint rotation speed reaches 25rad/s. Currently, STAR 1 has passed multiple product-level tests and reached the market application standards.
Relying on its self-developed modular design, the iteration speed of general-purpose robots has been further improved. In addition, "Star Motion Era" has recently launched the desktop-level robot M1 and the wheeled robot W1, which also verifies the key role that modularization plays in promoting the application of technology in different industries.
The participation of successive rounds of investors will also play a promoting role in the application of "Star Motion Era" in the field of embodied intelligence. For example, Alibaba will cooperate with "Star Motion Era" in fields such as AI technology and cloud computing; Lenovo Ventures, Zeyu Capital, Jinding Capital, Qingkong Tiancheng, and their affiliated institutions will also jointly explore the industrialization of humanoid robots in fields such as 3C electronics, kitchen appliances, home appliances, property management, industrial park operation, logistics, and ports with the company; Qingliu Capital and Yuanjing Capital will also assist the company in accelerating the commercialization process.
Chen Jianyu said that deep reinforcement learning represented by Alpha-GO and the large language model represented by ChatGPT are two important technical milestones for general-purpose humanoid robots. "In the next few years, these two technologies will be deeply integrated with humanoid robots and embodied intelligence. The field of embodied intelligence requires an original large-scale algorithm paradigm combined with more diverse and rich physical world data for training. It is believed that in the near future, the ChatGPT moment of robots will come."