Hard Krypton Exclusive | Tang Wenbin's Yuanlingli Acquires Logistics Robot Company and Secures Investments from Zhipu AI, SenseTime, Step Function, etc.

Author | Qiu Xiaofen

Editor | Yuan Silai

Yingke exclusively learned that the embodied intelligence company "ForceMind" recently completed a new round of financing. The investors are mainly several large model companies, including Zhipu, StepStar, and SenseTime. In addition, industrial investors such as Huqin and SAIC Hengxu continue to increase their investment .

"ForceMind" is a general embodied large model company founded in March 2025 by Tang Wenbin, the co-founder and CTO of Megvii Technology. The core founding members of the team are from the original Megvii Technology team.

Interestingly, this financing also marks a rare "reunion" of SenseTime and Megvii, two former competitors, in the wave of embodied intelligence.

In addition, including Alibaba, which exclusively led the A+ round of investment, this is also a rare occasion where four major domestic large model manufacturers have gathered in the embodied intelligence track. Previously, Zhipu only made small-scale investments in the field of embodied intelligence through the Z Fund, and StepStar has hardly made any investments in embodied intelligence.

This collective action also sends a signal that as the main battlefield of large model competition shifts from Token to Action, the embodied model with the ability to interact in the physical world has become the next high ground targeted by model companies.

Along with this financing, Tang Wenbin is also unifying forces and starting to integrate robot assets.

Yingke exclusively learned that "ForceMind" recently completed a merger with the logistics robot of "Atomix" (Force Aggregation) through equity mergers and acquisitions, aiming at the large-scale implementation and global expansion of embodied intelligence.

The business gene of "Atomix" can be traced back to 2016. At that time, Tang Wenbin led the intelligent logistics and robot scheduling business (Hetu System) within Megvii Technology, promoting multi-form logistics robot solutions.

Until July 2024, with the business changes of Megvii, Tang Wenbin separated the logistics robot business from the Megvii system and established "Atomix".

After several years of exploration, the sales volume of "Atomix"'s pallet four-way vehicles has reached the second place in the world, with more than 500 cumulative service projects. Customers include Uniqlo, Miss Fresh, and CATL. The company's annual revenue is nearly one billion.

As the hardware supply chain of embodied intelligence approaches maturity, the industry is approaching a high wall that must be crossed: the embodied brain. Compared with the clear evolutionary path of language models, the embodied intelligence model currently cannot even find low-cost, massive, and high-quality data, let alone find a convergent training paradigm. It can be said that the entire industry is groping in chaos.

In this situation, the integration of the body, the brain, and the data may become the new normal in the embodied track.

All along, the ideal state of the embodied intelligence industry is to build a real data flywheel. However, the actual situation is that the industry is in a "data deadlock" state. The model needs error data in real scenarios to promote evolution, but without a good model, robots cannot enter the scenario, and thus cannot collect real data.

Therefore, a source said that the merger of the two companies is essentially to close the loop between the model and the scenario and break the data deadlock.

As Tang Wenbin mentioned in a previous interview, Picking is the "atomic task" in the era of embodied intelligence. Picking is to embodied intelligence what Coding is to large models, and "Atomix" is like a continuously operating Picking data engine.

The robot of "ForceMind" making breakfast (Source/Enterprise)

It is understood that in the future, the real data generated by "Atomix" in more than 20 countries and over 500 projects will directly become the fuel for the model training of "ForceMind". And the embodied intelligence model trained by "ForceMind" will also quickly cooperate with the existing robots of "Atomix".

This idea may not be a castle in the air but is based on a certain technological foundation. Before this, "ForceMind" has launched the general embodied large model "DM0".

Tang Wenbin mentioned in a previous interview that at the data level, "ForceMind" has completed the first "big fusion of three types of data" in the industry - mixing and training Internet semantics, intelligent driving physical rules, and robot operation data to improve the scale and quality of data.

This cross-domain mixed training method enables "DM0" to break away from the dependence on specific hardware parameters. Like an experienced "old driver", it abstracts general physical laws from massive heterogeneous data and can migrate between robot bodies of various different configurations regardless of hardware differences, achieving general operation logic.

The robot of "ForceMind" making cocktails (Source/Enterprise)

More importantly, "ForceMind" also tries to extend the "thinking chain reasoning" of large models to the physical space. This enables "DM0" to achieve sub-millimeter-level precise operations with a small parameter scale of only 2.4B, and the success rate in long-range continuous tasks has been greatly improved.

Through a series of combined punches, "DM0" is trying to break the pain points of traditional embodied models, such as single data, inoperability after changing machines, and bloated parameters.

After this merger and financing, a strong player is emerging in the Chinese embodied intelligence industry. More importantly, this also indicates that the industry is entering the next stage - finding the scaling law of the embodied model.

This is by no means an obstacle that can be overcome by simply piling up robot bodies.

Just this week, the media reported that ByteDance is recruiting a person in charge of embodied intelligence technology with a large amount of money, targeting the core technical backbones of leading startups. At the same time, the overseas star embodied intelligence company Skild AI has just completed the acquisition of the robot automation business of Zebra (Zebra Technologies).

The trends of domestic and overseas giants are the same. As body manufacturers, data asset owners, model developers, and scenario operators start to gather at an accelerated pace, the industry has officially entered the deep water area.

Source of the homepage image | Internet

Typesetting | Fan Xinya

Welcome to communicate