Zhuoyu Releases Pansing Commercial Vehicle Smart Driving Solution and End-to-End Large Model for Heavy Trucks

Refuse to reinvent the wheel! From Chengxing to Panxing, Zhuoyu delivers the mass-production answer for commercial vehicle intelligent driving. Zhong Lilei, 13 hours ago. Only with an end-to-end large model can heavy truck advanced driver assistance be considered mature.

In today's intelligent driving industry, the passenger vehicle track has long been "rolled" to the point where not a blade of grass grows, with terms like end-to-end large models, world models, and BEVFormer dazzling the eyes, yet the commercial vehicle track—the hardest, most dangerous, and most essential—has remained strangely quiet.

In fact, heavy truck intelligent driving is currently in a delicate transition period, with the industry as a whole moving from "segmented modular" solutions toward end-to-end, but mainstream players still mainly follow the modular BEV route. Everyone knows end-to-end is the inevitable choice for intelligent driving technology, but no one wants to be the first company to put end-to-end onto heavy trucks.

On October 29, Zhuoyu held a commercial vehicle technology launch event at the Zhuhai International Airshow Center, bringing the world's first mass-produced commercial heavy truck end-to-end large model, as well as the commercial vehicle advanced driver assistance solution built on that model—Panxing. Zhuoyu, the first to eat the crab.

Passenger vehicle end-to-end large models have long been popularized on a large scale, so why are there still very few in the heavy truck field? The answer is not that no one wants to do it, but that the heavy truck scenario magnifies the difficulty of end-to-end several times over.

Whether passenger vehicles or commercial vehicles, the prerequisite for training an end-to-end large model is having sufficient data resources. According to information released by the Ministry of Public Security, as of June 2026, China's vehicle parc was about 370 million, while the heavy truck parc was only 8.6 million.

Passenger vehicles can accumulate fleet road-test data at the hundred-million-kilometer level in a year, with low collection costs and mature channels, while heavy truck data is highly closed, difficult to collect, and public datasets are scarce, so high-quality samples usable for training are naturally few. Without enough data, even the most advanced algorithms can only be castles in the air—this is the first hurdle that heavy truck end-to-end cannot bypass.

Second, heavy trucks have the nickname "hundred-ton kings," easily exceeding 30 tons when fully loaded. The heavier the vehicle, the greater the inertia, and the braking distance needed grows accordingly; moreover, heavy trucks have long braking chains, naturally entailing a delay of nearly 1 second. Add to that the trailing attitude of trailers, braking heat fade on long downgrades, and the effect of centrifugal force in curves on handling, and the control difficulty of heavy trucks far exceeds that of passenger vehicles.

That is to say, when an obstacle exists at the same distance ahead, a heavy truck needs to brake earlier than a passenger vehicle to ensure no collision occurs. At the hardware level, a heavy truck's intelligent driving system must have perception capability beyond that of passenger vehicles in order to obtain information about obstacles ahead in advance and take measures such as avoidance and braking.

For an ordinary passenger vehicle, the 100 km/h-to-0 braking distance is generally within 50 meters; considering factors such as highway speed limits and the reaction speed of the intelligent driving system and braking chain, a sensor perception distance of about 200 meters is enough. On a "hundred-ton king" heavy truck, this number must at least double, and it must be able to identify small, irregularly shaped, low-reflection obstacles in complex lighting, dust, rain, and fog environments, so that the "slow-reacting" heavy truck can brake in time.

Ultimately, scarce training data and high perception distance requirements are both related to safety. Beyond safety, heavy truck users also have higher cost-control requirements than passenger vehicle users. Heavy trucks often consume dozens of liters of fuel per 100 kilometers, and logistics vehicles frequently travel hundreds to thousands of kilometers; "seven or eight yuan per brake tap" is no joke. Even a slight fluctuation in fuel consumption will affect operating costs. In the past, compared with intelligent driving systems, many veteran drivers trusted their own "footwork" more.

It is precisely because of these difficulties that intelligent driving companies are relatively cautious in their attitude toward commercial heavy trucks, and without sufficient confidence, they dare not rashly launch solutions based entirely on end-to-end technology. As for why Zhuoyu dares to be the first company to eat the crab, the October 29 launch event explained the reasons thoroughly. To let heavy trucks use end-to-end, Zhuoyu has given its own solution.

At the end of 2025, Zhuoyu proposed the "mobile intelligent base" concept, planning to build a capability foundation supporting multiple vertical businesses such as passenger vehicles, commercial vehicles, unmanned logistics, and future mobility. In April 2026, Zhuoyu released the "intelligence for all mobility" strategy, promoting physical AI from "one vehicle, one intelligence" to "all things sharing one intelligence."

Panxing is precisely the solution Zhuoyu built based on this strategy. It reuses the foundation model and AI base already mass-production-validated in Zhuoyu's passenger vehicles, and, based on the experience of more than 3 billion kilometers of passenger vehicle veteran driving, conducts reinforcement training with up to 400 million frames of heavy-truck-specific data, enabling the model to deeply learn commercial-vehicle-specific operating conditions such as trunk-line logistics, highway transport, complex traffic flows, ultra-long braking distances, and trailer attitude changes.

A unified underlying architecture means that the algorithm iteration, safety validation, and mass-production experience Zhuoyu has accumulated in the passenger vehicle field over the years can be migrated to commercial vehicles at low cost. Only with the general "foundation" paired with heavy truck "ingredients" can a truly suitable end-to-end large model for commercial vehicles be trained. Under the premise of sharing the same base model, Panxing performs scenario-based adaptation for different commercial vehicle carriers, balancing mass-production efficiency and scenario depth.

Zhuoyu did not build a new intelligent driving system from scratch, but instead used the passenger vehicle platform foundation model for migration, used the passenger vehicle platform foundation model for migration, to make the model understand heavy trucks better; moreover, the data generated from heavy truck mass production will further feed back into the platform foundation model, enhancing model capability and reducing the cost of cross-vertical applications. So, let the same "brain" learn to drive commercial heavy trucks!

Considering the software and hardware costs of intelligent driving technology, Panxing is divided into two versions, PDS 4 and PDS 5. PDS 4 corresponds to the highly perceptive large model 4.0 and above, and supports highway NOA; PDS 5 corresponds to the native multimodal model, supports full-scenario capabilities of highway NOA and urban NOA, and can be further reused in buses, light trucks, and other models that often drive in urban areas.

The data collected by these vehicles will feed back into the foundation model, driving the continuous evolution of the end-to-end large model.

To further ensure vehicle safety, on the hardware side, the Panxing solution is the first to feature the Jimu 2.0 perception system and adopts an in-cabin LiDAR solution, which is unafraid of rain, snow, and other severe weather and does not require separate cleaning equipment. Even if hit by splashing small stones, the sensor is protected by the front windshield and is not easily damaged.

At high speeds, this perception system still has a detection range of up to 500 meters and can detect obstacles that are small in size, irregular in shape, and low in reflectivity, such as black tires or fallen cardboard boxes. When faced with interference factors such as strong light, rain and fog, and dirt, the algorithm also automatically filters noise to ensure the accuracy of the system's judgment.

In addition, compared with passenger cars, heavy trucks also have the disadvantage of large blind spots. Imagine sitting in a heavy truck more than ten meters long with an extremely high cab; sometimes it is difficult for us to promptly spot pedestrians, non-motorized vehicles, and small vehicles crossing ahead or to the side. Zhuoyu Panxing's multiple sensors distributed across the vehicle body, combined with software algorithms, can estimate the angle between the tractor and trailer and the vehicle body posture, and dynamically predict collision risks through real-time load perception.

This intelligent assisted driving system can also use dynamic fine-tuning algorithms to allocate power in advance, achieve precise control of the vehicle body, and further reduce the possibility of collisions.

Based on the Jimu 2.0 perception hardware and dynamic fine-tuning software algorithms, Zhuoyu Panxing has created a full-dimensional safety system. This system has three layers. The first layer is multi-scenario safety reminders, which can use sensors to monitor pedestrians or vehicles ahead and to the side and issue warnings to the driver in advance.

If the driver does not react quickly, the second layer of the full-dimensional safety system, active safety, will come into play and proactively take avoidance or braking measures to avert risks. The third layer, multiple safety redundancies, supports fault self-diagnosis and minimum risk strategies, and can still handle problems with the safest solution when they occur.

To address energy consumption, Panxing's debut AI reinforcement learning fuel-saving algorithm can adjust vehicle speed and power distribution in advance based on the slope and distance of the driving section and through dynamic fine-tuning algorithms, combining the experience of veteran drivers with precision that veteran drivers do not have. According to Zhuoyu's actual measurements, Panxing can effectively reduce the fuel consumption of heavy trucks by 4.5%-7.5%, saving about 20,000 to 30,000 yuan in fuel costs per year. The cost spent by enterprises or individuals on the software and hardware of the Panxing intelligent driving system can be earned back in just one year of driving a heavy truck from the fuel savings alone.

Zhuoyu's core method for solving the difficulty of deploying end-to-end large models in the heavy truck field is to reuse mature technologies from the Chengxing platform on commercial heavy trucks. Through targeted data training, a dedicated perception system, and targeted calibration, it matches the characteristics of heavy trucks, such as difficult handling and long braking distances. The AI reinforcement learning fuel-saving algorithm further solves the fuel cost problem.

According to Zhuoyu, China's top six commercial heavy truck brands—XCMG Group, Shacman Heavy Truck, Sinotruk, FAW Jiefang, Foton Motor, and Dongfeng Commercial Vehicle—have all reached cooperation with Zhuoyu. The mass-produced and deployed Shacman Delong X5000 further proves the mass-production feasibility of Panxing. The top six commercial heavy truck brands have a market share of about 93%, which means that the commercial heavy truck brands we often see have all joined the Zhuoyu Panxing camp. Not only heavy trucks: the era of all things sharing one intelligence is coming

Panxing has verified the cross-vertical replication capability of the "mobile intelligent base" in commercial vehicle scenarios. It proves that what Zhuoyu sells is no longer just an intelligent driving solution for a certain vehicle model, but a capability base that can be repeatedly replicated and infinitely expanded. From passenger cars to heavy trucks, from buses to logistics vehicles, Zhuoyu uses the same base to carry all things.

At the launch event, Zhuoyu CEO Shen Shaojie also shared that leading domestic bus company Yutong Group has reached cooperation with Zhuoyu, and the Yuwei E9 PRO will be the first to be equipped with Panxing™ PDS 5.

For the industry, Panxing is more like a clear signal. Once the first company bold enough to reuse a passenger car intelligent driving base for commercial vehicles has succeeded, other autonomous driving companies and automakers will most likely follow suit. This track of commercial vehicles, the toughest and riskiest and also the one that most emphasizes input-output returns, may thus usher in its own era of intelligent driving popularization.

While the entire industry is still struggling with "whether to do it," Zhuoyu has already given its own answer with "from Chengxing to Panxing." True masters never start everything from scratch. Looking back suddenly, the technology is right there in the dim light.

Source: Leikeji

The images in this article are from: 123RF licensed image library. Infringement report Return to sharing Related articles All comments