Author | Zhang Lianyi, Editor | Jiang Tianzhuang
Qipanjing Town is located in the west of Ordos, Inner Mongolia, with a permanent population of less than 100,000. More than 20 years ago, a group of builders came here and built a power plant and transported coal on a grassland where not even a single restaurant or hotel could be found, successfully establishing an industrial park with an output value of CNY 100 billion.
On September 22, KargoBot held its launch event here once again. After taking the stage, CEO Wei Junqing first shared this story. He said, "Five years ago, KargoBot came here, hoping to use AI (Artificial Intelligence) technology to create another breakthrough from 0 to 1."
This goal has already yielded phased results. The company has achieved a closed loop for autonomous freight transport in Qipanjing, and will achieve a positive Unit Economics (UE) model at the route level in 2025...
This time, Wei Junqing announced taking another major step forward: launching a scaled AI transport network.
During the speech and the post-launch interview, he further broke down the underlying logic: why KargoBot dares to bet on physical AI in the grueling freight industry, why it believes this network can grow on its own, and where the people who used to hold the steering wheel will go as autonomous fleets increase.
01. Choose the Right Battlefield First, Then Talk About the Network
Wei Junqing has a simple classification for physical AI: one type revolves around serving people, and the other revolves around cargo production. KargoBot has firmly chosen the latter.
"Production is an absolute rigid demand and will also be the first to achieve commercialization and scaling in the AI field." He gave an example: everyone chats with GPT, but ultimately, the most money spent on large models is on using them to make PPTs, write reports, and write code. "Our company's current resource investment in AI code generation is two orders of magnitude higher than that in chat functions."
Wei Junqing's another judgment on robotics is that it must evolve from single-agent intelligence to economies of scale, shifting from competing on which robot runs faster, smarter, or does housework better, to focusing on economies of scale, namely mass production and cost reduction. Moreover, the ultimate winner must possess network effects to be accepted and used by everyone.
Based on this, KargoBot mainly focuses on two things: deeply cultivating industrial production scenarios, focusing on embodied intelligence and physical AI around bots; and going all-in on cargo intelligence around network effects.
After the battlefield was selected, the product line gradually took shape. Currently, KargoBot has three product lines: the hybrid intelligent platoon has been deployed on a large scale, increasing the transport profit margin from the traditional model to 10%-18%; single-vehicle intelligence targets short-distance, parks, and ports, with the full-cycle transport profit margin increasing by over 20%; and the cabin-less transport robot is in the testing and iteration stage, with the economic benefit per vehicle expected to increase from 20% to 30%.
Where does the profit come from? During the interview, Wei Junqing broke down the cost structure. The traditional cost in the freight industry is "33211", namely 30% for tolls, 30% for energy, 20% for labor, and 10% for vehicle maintenance and depreciation. In recent years, through new energy transition, energy costs have decreased, but labor costs have risen. The platoon mode allows a single individual to create greater value.
"By having one person drive two vehicles, the proportion of labor costs drops by nearly half; having one person drive three vehicles can further reduce it," Wei Junqing stated. "Although some costs have increased, such as sensors, the overall positive economic model has been achieved, showing further improvement compared to the traditional model."
After the product is proven, the next step is network expansion. KargoBot's path is the "1+N+X strategy": "1" represents the successful model in Ordos, "N" represents regional expansion, and "X" represents scenario expansion.
However, he does not plan to build the network alone. "We prefer to be an enabler for network construction." The company has cooperated with 8 OEMs and 30 ecosystem partners, covering regions such as Inner Mongolia, Ningxia, Gansu, Xinjiang, Shaanxi, Shanxi, and Hebei, and has been implemented in scenarios including steel, express delivery, industrial products, components, agriculture, and animal husbandry.
Once the AI transport network matures in the future, Wei Junqing stated that it might also be opened to individual drivers or small fleets. "We are also very willing for them to participate. We believe this is definitely possible and is what we hope to do."
Of course, network effects mean larger scale, which also implies greater challenges.
Wei Junqing agrees that the challenges of fleet scaling lie not only in technology but also in after-sales, operations and maintenance (O&M), dispatch, and safety assurance.
KargoBot's response strategy is as follows:
Proactive product design consideration: Both hardware and software iterations undergo full life-cycle analysis, incorporating O&M, quality assurance, and after-sales difficulties in advance to reduce the investment in spare parts and operations;
Intelligent dispatch and logistics forecasting: Relying on the operation of hundreds of vehicles, dispatch algorithms are accumulated on the KargoBot platform. Combined with big data to predict logistics demand, vehicles are intelligently dispatched in advance according to route cargo sources and seasonal and periodic changes;
Systematic safety and service assurance: As the vehicle scale expands, the focus is on ensuring operational safety and transport services, treating safety as a bottom-line issue. Mobile service stations and mobile safety assurance systems are gradually established, forming a network through route patrols and stations to support large-scale networked operations.
As for whether they worry about automakers directly entering the field in the future, KargoBot's answer is no.
"Automakers hope to build a complete product, with both a body and a soul. But over the years, most automakers, especially commercial vehicle manufacturers, have encountered many difficulties in developing high-level autonomous driving software." In Wei Junqing's view, automakers' product development is plan-driven project management, while autonomous driving software is continuously fed and iteratively upgraded through data, with delivery being just the starting point. It is difficult to integrate two completely different product cultures.
He emphasized that KargoBot has established deep connections with major domestic commercial vehicle companies. "Under the definition of our software, we enhance capabilities such as chassis drive-by-wire, achieving deep integration on both sides to make the vehicles more competitive."
02. Fourfold Positive Feedback, the Network Grows on Its Own
Why can this network be built and accelerated?
Wei Junqing's answer is "self-growth": it does not rely purely on external forces to pile up nodes, but on the internal positive feedback of the autonomous driving network.
The more data, the smarter the Agent. This is the first positive feedback.
"In 2021, KargoBot pioneered the hybrid intelligent platoon mode, tapping into 4 million edge case data points. These scenarios have a low probability of occurring in the real world, cannot be replicated in simulations, and are difficult to encounter in testing," Wei Junqing introduced. "For example, temporary signs at toll stations, animals crossing, icy roads, and strong crosswinds on mountain roads."
In his view, KargoBot possesses a fleet of hundreds of autonomous trucks and tens of millions of kilometers of L4 autonomous driving operational mileage. This gives it a leading and genuinely effective database, from which the "Truck Foundation Model" has been trained. This model can be deployed and operated in various scenarios and different routes, such as national highways, expressways, ports, railways, and mining areas, under platoon and single-vehicle modes.
"These real-world data have given us a deeper understanding of heavy truck characteristics and physics. Relying on the accumulation of real operational data, our key takeover metric data has improved by more than 180 times over the past two years," Wei Junqing gave an example. When multiple vehicles follow at high speed in a platoon, if the leading vehicle presses the accelerator or brake slightly more, it will be amplified by the following vehicles, potentially leading to fuel consumption higher than human driving, adding several thousand CNY in costs annually. "For such problems, our engineers were completely at a loss initially, but now they handle them with ease."
The deeper the scenario, the stronger the economics. This is the second positive feedback.
KargoBot has an internal calculator that computes the full life-cycle value and cost for customers in every transport trip. While others haggle over every penny, they calculate the "cost per ton-kilometer", which has led to the formation of the aforementioned product lines.
Wei Junqing pointed out that since last year, KargoBot has integrated the factory demand system and coal mine production system of Ordos Group, as well as the entire process of entering and exiting the factory, loading and unloading, without the participation of safety officers throughout, becoming the world's first real normalized unmanned operation closed loop with UE economics.
The more partners, the more efficient the network expansion. This is the third positive feedback.
Compared to the very beginning, when deploying a route required many engineers to invest 8 months to 1 year, the time has now been reduced by nearly half, and it is still evolving at a faster pace every year.
The more perfect the compliance system, the thicker the foundation of safety trust. This is the fourth positive feedback.
"KargoBot has obtained the most L4 autonomous driving truck commercial unmanned operation licenses in China, possessing the first nationwide and the only one in Inner Mongolia cross-city commercial pilot qualification, enabling the autonomous driving industry to achieve the leap from city-level to provincial-level for the first time," Wei Junqing stated. The first principle of the company's cultural values is "no safety, no everything". Since the implementation of operations, no safety liability accidents have occurred.
These fourfold positive feedbacks have enabled KargoBot to form a self-growing transport robot network based on the three-layer architecture of "AI + Robot + Solution", Wei Junqing summarized.
03. From Behind the Steering Wheel to in Front of the Screen
Technological progress has never been about "eliminating jobs", but rather "redefining" them. However, for specific individuals, there is indeed a period of pain and a skill threshold.
KargoBot's platoon mode started as early as 2021. The lead vehicle is equipped with L2 intelligent driving functions and requires a driver, while the following vehicles are all unmanned vehicles with L4 autonomous driving capabilities.
Under this mode, the driver's role begins to change. As an AI navigator, which is not entirely the same as an ordinary driver, he needs to drive the vehicle while taking care of the following vehicles, constantly monitoring the fleet status on the small HMI (Human-Machine Interface) screen while driving.
However, the truly systematic transformation is the "Wrangler Plan" launched this year, namely the "Driver Transformation Project". It mainly selects experienced drivers from partner fleets and certifies them as autonomous driving platoon managers, helping drivers transition from "people holding the steering wheel" to "people managing the fleet".
"We will conduct specialized training, from theoretical assessments and practical assessments to closed training, to help drivers quickly integrate into the role of AI navigators," it is reported. KargoBot uses mileage and accuracy as assessment criteria, and those who fail to meet the standards will undergo stricter follow-vehicle training.
At this launch event, five representatives of AI navigators who completed the standardized training took the stage to receive their graduation certificates.
In fact, such training is not one-way. Wei Junqing stated that drivers are very familiar with the characteristics of the vehicles and scenario needs, and often provide feedback for each version, such as the optimal solution for following distance and braking speed, which is very helpful for autonomous driving model training. "Rather than saying we are training them, it is more about empowering them. Their operational skills are already very high, and they are very serious and responsible, and they are also very eager to participate in new things."
More effort also means more rewards. According to Wei Junqing, the salary of AI navigators will increase to a certain extent compared to ordinary truck drivers.
Starting from Qipanjing, KargoBot is building a self-growing transport robot network. The more nodes in this network, the faster and more efficient the expansion speed. And at every node, AI navigators move from behind the steering wheel to in front of the screen, becoming a key link in the transition of unmanned freight from testing to operation.
-END-