① Since its establishment, in less than a year and a half, the variable Robotics has completed seven rounds of financing, with a total financing amount exceeding 1 billion yuan; ② This round of financing was exclusively invested by Meituan, indicating that major industry players are accelerating their layout in the areas of embodied brain and application landing.
On May 15, the Star Daily reported (Reporter Li Mingming) that the embodied intelligence track has welcomed another round of financing.
Recently, the variable Robotics announced the completion of several hundred million yuan in Series A financing, led by Meituan's strategic investment and followed by Meituan Longzhu. The company stated that this round of financing will be used to continuously accelerate the synchronous iteration of fully self-developed end-to-end general embodied intelligence large models and the robot Ontology, as well as collaboration and landing of intelligent solutions in multiple future application scenarios.
The reporter from the Star Daily noticed that since its establishment, in less than a year and a half, the variable Robotics has completed seven rounds of financing, with a total financing amount exceeding 1 billion yuan.
When discussing why the company insists on the "end-to-end large model integrating small and large brains" approach, the founder and CEO of variable Robotics, Wang Qian, explained to the reporter of the Star Daily that the WALL-A model from the company's independently developed Great Wall operation large model series possesses multimodal information fusion capability, able to integrate visual, language, and motion control signals to achieve a complete closed loop from perception to execution. Unlike the "layered and step-by-step" architecture, WALL-A is an "end-to-end unified model." "WALL-A has demonstrated zero-sample generalization capability in multiple previously unseen physical scenarios, which is one of the key thresholds for general robotics capabilities, especially showing significant advantages in long sequences and complex motion control."
He also revealed that commercial Retirement is an important direction targeted by variable Robotics.
This round of financing was exclusively invested by Meituan, which means that on the embodied brain side and application landing, industry giants are accelerating their layout.
Exclusively invested by Meituan.
The Variable Robot was established in December 2023, focusing on the research and development of the 'General Embodied Large Model', using real-world data as the primary data source to build a general-purpose robot capable of fine operations.
The Variable Robot has currently completed seven rounds of financing, gathering several institutions including Lightspeed China Partners, Legend Capital, Huaying Capital, Yunqi Capital, and GF Xinde.
This round of exclusive investor Meituan has a broad layout in the field of robots. As early as 2017, Meituan began to venture into applications such as Drones and hotel robots; in 2021, Meituan officially proposed the 'Retail + Technology' strategy to continuously layout the robot field, focusing on accelerating the unmanned process in service scenarios and enhancing delivery and operational efficiency; in December 2022, Meituan established a Robot Research Institute, adopting a self-research + investment approach, investing in several companies such as Liao Biao Robot, Future Robot, and Nine Senses Intelligence, covering fields like Logistics Automation, humanoid robots, and last-mile delivery.
In terms of application scene implementation, Meituan has made many attempts. For example, in September 2024, Meituan signed a strategic cooperation agreement with Galaxy General Robot to jointly create the world's first humanoid robot intelligent pharmacy solution. Currently, there are two pharmacies in Peking in pilot operation.
In food delivery, Meituan collaborated with Yunji Technology to use robots to solve the 'last 100 meters' problem in food delivery.
It can be seen that Meituan's investment strategy is dual: it hopes to reduce risks through investment while achieving more possibilities by collaborating with companies on different technology paths.
An investor associated with the robot industry chain told the Star Daily reporters that Meituan has invested in robot companies with different technology paths, covering Hardware, Software, algorithms, etc. The intention is to build a full-stack technology system from perception, decision-making to execution, and achieve deep integration of robot technology with its own Business to enhance service efficiency and user experience. In the case where the technology in the robot industry has not yet unified, investments are used to empower the ecosystem and add momentum for future development.
To create an 'Embodied Brain'.
Currently, the Robot Industry is gradually evolving into more specialized segments, with startups in the field of Asia Vets Intelligence pursuing various pathways such as focusing on Ontology, developing brains, and whole machines.
"Ontology companies" are favored by capital, while "embodied brains" are gradually gaining more attention in the market. The fundamental reason is that the development of "embodied brains" in the country is limited, lacking a product-level "embodied brain" aimed at the Robot Industry. The absence of brain capabilities severely restricts the autonomous decision-making intelligence of robots.
As one of the earliest companies in the country to achieve a generalized embodied intelligent large model using a fully end-to-end path, Variable Robot is committed to building an integrated system architecture for unified perception, planning, and control.
Wang Qian introduced that there are still many unexplored areas in the field of embodied intelligent large models. In China, the combination of the first batch of hundred-billion-level large language models and robots is still shallow, often limited to simple voice interactions and perception planning. Meanwhile, globally, there has yet to be a universal large model capable of truly solving complex operational problems in the physical world. Traditional robots are typically based on specific scenarios and tasks, making it difficult to autonomously adjust strategies based on changes in the environment and tasks.
Therefore, Variable Robot trains a highly generalized large model of embodied intelligence, referred to as the unified large model, connecting embodied intelligence to a common underlying model.
It is reported that this model achieves a longitudinal unification from end to end. From the most basic video and Sensor signals to the final robot's speed, posture, and torque, it is completely solved using a single model; the company can achieve generalization and transfer of various physical environmental variables and action patterns with very few samples, while also having certain advantages in complex operations across long sequences; in terms of task complexity, it can perform detailed, random tasks involving complex topological structures such as pulling zippers, buttoning, and organizing clothes, demonstrating the ability for precise operation and handling complex environments.
At the Hardware level, the company independently develops and continuously optimizes robots’ Ontology to adapt to multi-modal large model control, better meeting the needs for precise operation and stable running in open environments. Currently, Variable Robot's Ontology has been applied in scenarios involving multi-step complex tasks.
Wang Qian believes that the development of embodied intelligence in the short term relies on model algorithm advantages, in the medium term on data advantages, and in the long term, the core lies in product advantages. Variable Robot deepens its accumulation around these three major directions, exploring service closed loops in different scenarios.
Wang Qian graduated with a bachelor's and master's degree from Tsinghua University, then went to the USA to study for a Ph.D. at the University of Southern California. In December 2023, he returned to China and founded Variable Robot in Shenzhen, serving as the company's founder and CEO.
According to incomplete statistics from the Star Daily, companies in the same field as Variable Robot are spread across both domestic and international markets, with international companies including Skild AI, Google's DeepMind, and Physical Intelligence (PI). These companies are actively laying out in the embodied intelligent large model field, adopting an end-to-end approach for technology research and development.
Mainly domestic companies include Qianjue Technology, which is committed to creating embodied intelligent large models and providing "one brain, multiple forms" solutions for traditional robot companies; Zhihui Square, which has signed contracts with several leading domestic and international automotive manufacturers, high-end manufacturing enterprises, and Internet giants, and has made certain progress in the research and commercialization of embodied large models; and Qiongche Intelligence, which is also focused on the research and development of embodied intelligent brains.
