Former Li Auto executives are applying the "technology philosophy" from the car - making industry to robots and have raised 2 billion yuan in half a year.

In March 2026, a company that had been established for only eight months managed to squeeze onto the most crowded table in the embodied intelligence track with five rounds of financing, 2 billion RMB, and a valuation of one billion US dollars.

If you look at the list of investors in Zhijian Dynamics, Sequoia, Legend Capital, BlueRun Ventures, and Yuanjing Capital, along with the two strategic investors Tencent and Alibaba, it's a luxurious lineup in any track.

However, what's more worth noting than the financing is the company's founding team - Wang Kai, the former CTO of Li Auto, Jia Peng, the person in charge of intelligent driving technology R & D, and Wang Jiajia, the person in charge of intelligent driving mass production. This is almost half of the core intelligent driving team of Li Auto.

The question is: Can these people who have experienced intense competition, fought battles, and proven themselves in the car - making field now apply Li Auto's technical methodology to robots and make it work?

Cai Neng Quan's judgment is: Zhijian Dynamics is not copying Li Auto's technology but replicating Li Auto's technical concept - that systematic combat ability from 0 to 1 and from 1 to scale. This is the real reason why VCs are willing to continuously increase their investment.

Remove the VLA model from the car and install it in the robot's body

Jia Peng did something at Li Auto: He led the R & D and delivery of the world's first VLM + end - to - end fast - slow dual system and was also the person in charge of the R & D of the first VLA model. VLA, short for Vision - Language - Action, is a vision - language - action model. In essence, it is the "brain" that enables machines to understand the world, understand human language, and then perform tasks.

When this technology is applied to cars, it solves the problem of autonomous driving; when applied to robots, it solves the same problem of interacting with the physical world. The only difference is that cars run on four wheels on the ground, while robots move in space with two arms or two legs.

What Zhijian Dynamics is doing now is to remove the VLA model from the car and install it in the robot's body. Their LaST₀ base model combines the world model's understanding and prediction of the physical world with the fast - slow thinking of VLA. Translated into plain language: It enables robots to react quickly and also slow down to think about complex problems. This architecture is in line with the VLM + end - to - end fast - slow dual system that Jia Peng developed at Li Auto in terms of technical logic.

Another interesting detail: The ManualVLA long - range task model released by Zhijian Dynamics has had its related papers accepted by CVPR 2026. CVPR is a top - tier conference in the field of computer vision. Being able to publish papers at this level of conference shows that the company has a solid technical foundation.

However, this approach comes at a cost: high training difficulty, high computing power consumption, and a large demand for data. Zhijian Dynamics' solution is borrowed from Li Auto - only develop a general ontology to cover as many scenarios as possible and improve data universality and reuse rate. This logic has been proven in the car industry: Li Auto only makes one model to dominate the market and reduces costs through economies of scale. Now, when applied to robots, the idea is exactly the same.

The "shadow mode" is standard in the intelligent driving industry, but no one in the robot industry has done it yet

The data closed - loop is the lifeline of embodied intelligence. For robots to work in the real world, they must have the ability to learn while working.

Zhijian Dynamics' solution is to deploy and pre - embed additional computing power at the edge side, enabling robots to complete data collection, training, testing, and verification on the ontology. This paradigm has a specific name in the intelligent driving industry: the shadow mode.

The shadow mode was first invented by Tesla. When the car is on the road, the intelligent driving system silently observes. If the driver's operation is inconsistent with the system's decision, the data is sent back to train the model. The advantage is that it can collect data at low cost in real scenarios without users paying and without special testing.

Wang Jiajia worked at Li Auto for several years and led the mass production and delivery of the end - to - end large model. He joined Li Auto in 2021 and promoted the intelligent driving system onto the car all the way from the map - based to the map - less solution and then to the end - to - end large model. In August 2025, after the delivery of Li Auto's latest VLA large model project was completed, he left and joined Zhijian Dynamics. Now he has brought this mass production and delivery experience to the robot industry.

Zhijian Dynamics' proposed "Human data is all you need" robot learning paradigm divides the learning process into three stages: In the pre - training stage, a large amount of operation data is collected manually to improve generalization ability; in the downstream task stage, human demonstrations are used to quickly expand task exploration; in the post - training stage, real - time manual guidance is used to participate in online learning. This hierarchical and progressive training idea is exactly the same as the data closed - loop logic in the intelligent driving industry.

But the problem lies here: The driving data of cars is relatively standardized, with clear rules for road conditions, lanes, and traffic signals. The operation data of robots in factory, supermarket, and logistics environments is several orders of magnitude more complex and random. Whether the shadow mode can work on robots after working on cars remains to be verified in real scenarios.

Build the first - generation machine in 45 days and expand to three locations in half a year. This rhythm was developed during the Li Auto era

Zhijian Dynamics has a very tight execution rhythm.

It took less than 45 days from the arrival of the first employee to the birth of the self - developed first - generation ontology machine. So far, two generations of ontologies for B - end and C - end users have been developed, small - batch production has been achieved, and PoC verification has been fully launched. The company has completed its layout in Beijing, Shanghai, and Suzhou, and the Global Innovation Center in Suzhou has been established.

This sense of rhythm was developed during the Li Auto era. When Wang Kai served as the CTO of Li Auto in 2021, he led the team to set an industry record of achieving mass production of the intelligent driving system on the car in seven months. At that time, the competition in the new - energy vehicle manufacturing industry had reached a stage where progress was measured in months. Whoever was one step slower might fall behind.

During his five - year tenure at Li Auto, Jia Peng experienced the iterative process from BEV perception to AD Max 3.0 and then to the VLA model. This high - intensity and fast - paced technical iteration experience is also applicable to the robot track. Although the current embodied intelligence industry is not as competitive as the car industry, the financing rhythm, product iteration, and scenario verification are all accelerating.

Zhijian Dynamics' strategy is clear: Start with closed - loop scenarios. Factory workshops, supermarkets, and logistics are the first targets. From closed to semi - open and then to fully open scenarios, from B - end to C - end users, and from the domestic to the overseas market. This progressive path is almost the same as the evolution logic of intelligent driving from highway NOA to urban NOA and then to full - scenario applications.

Jia Peng said, "Systematization is our difference." It may sound like a slogan, but in the embodied intelligence track, it is indeed a real threshold.

What today's robot companies lack is not demos but the systematic ability to develop products, put them into real - world scenarios to work, and keep them running stably. The Zhijian team has run this system during the Li Auto era and is now running it again in a different track.

With five rounds of financing in half a year, 2 billion RMB, and a valuation of one billion US dollars, Zhijian Dynamics has rushed into the unicorn club at the fastest speed.

But this is just the beginning. No matter how attractive the technical route is, ultimately, it depends on whether the robots can work stably in real factories, whether customers are willing to pay, and whether the data closed - loop can really work. In the embodied intelligence track, there is no shortage of news about fundraising, but there is a shortage of companies that can survive, scale up, and make money.

The Zhijian team has witnessed the most brutal elimination rounds and endured the darkest moments during the Li Auto era. Now, in a different track, they still use the same strategy: apply the technical methodology verified in the car industry to robots, and then compete in execution, rhythm, and system.

The VCs have already placed their bets. Now it's time to see if these "Li Auto people" can tell the same story with robots. Technology can be reused, but the real test has never been in the laboratory.

This article is from the WeChat official account "Cai Neng Quan", written by Sijiali and reprinted by 36Kr with permission.