China's smart trucks are "way ahead": Truck NOA has been implemented for 50 million kilometers with zero accidents, the world's first

Posted by Jia Haonan from the Assistant Driving Temple
Smart Car Reference | Public Account AI4Auto

The mature "Optimus Prime" can already go out and make money on his own.

At this moment, there are smart heavy truck carriers on almost all major freight trunk lines in East China, North China, South China, Northwest China...

On the highway section, the truck's intelligent driving system is fully responsible for the driving tasks, autonomously controlling the accelerator, brakes, autonomous lane keeping, and automatic cruise:

And change lanes autonomously or with a lever based on real-time traffic conditions:

It has basically the same function as "pilot assisted driving" for passenger cars, but serves the core demands of the logistics industry - safety and cost reduction.

The truck's NOA function is equipped with Inceptio Technology's Xuanyuan intelligent driving system .

In fact, Inceptio Trucks NOA has been in commercial operation for a year and a half, and has accumulated more than 50 million kilometers of mileage and has a record of zero accidents and zero insurance claims under intelligent driving conditions .

At the same time, the average fuel saving in operation is 2-10%, and the average labor cost saving is 20-50%.

The absolute mileage of commercialization and the depth of exploration in the implementation model are unique in the global smart truck circuit.

In the autonomous driving landscape, Robotaxi shows a situation where China and other countries are chasing each other. In the field of commercial vehicles, the latest developments show that China has become the high ground and center.

Truck NOA?

"NOA" is a very popular concept in the passenger car field in the past two years. The accurate name should be "navigation assisted driving".

Compared with ordinary L2 assistance in the past, the essential improvement and difference of NOA is that it can independently choose travel strategies based on navigation routes at a macro level, and respond to various scenarios independently based on real-time road conditions at the micro level.

For example, when to change lanes, following distance, getting on and off highway ramps, etc., no longer require direct operation by human drivers, and the user's role has also changed from driver to safety officer.

The value of NOA is reflected in this: greatly reducing driver fatigue and burden.

As for "Truck NOA", its main functions and scenarios are similar to those of passenger cars: it currently realizes L3 intelligent driving on highways.

In addition to reducing driver burden, truck NOA also has new functions and significance, and is also the main starting point for commercialization: handing over the driving task to the system will achieve better performance in terms of energy consumption and safety of the vehicle.

Specifically, the functions implemented by Inceptio Trucks NOA include:

Automatic cruise , including adaptive cruise, vehicle distance/vehicle speed adjustment... Its function is that under high-speed road conditions, the system can travel independently according to the navigation route.

Intelligent lateral control , including intelligent lane changing, can use the lever to issue lane change instructions, and can also change lanes independently, in addition to lane keeping, intelligent avoidance, etc.

Intelligent fuel saving , by analyzing the navigation route, first planning the global vehicle speed, and then dynamically adjusting the vehicle speed according to the implementation situation, "learning from" excellent human truck drivers, try to maintain a constant speed throughout the journey and avoid sudden braking and acceleration.

Finally, there are safety functions , including AEB mandated by regulations, as well as active safety functions integrated with Inceptio Automotive Cloud.

What supports the NOA function of trucks is actually highly similar to the intelligent driving solution of passenger cars.

1-3 lidars , 5 millimeter-wave radars, 7 8-megapixel surround-view cameras, and underlying computing power support, with AI computing power up to 256TOPS, which can be flexibly configured.

In fact, from the perspective of implemented functions, Inceptio’s truck NOA solution is not considered “heavy”. On the contrary, in terms of the number of lidars and underlying computing power, it is much “lighter” compared with the passenger vehicle NOA (passenger vehicle NOA solution) . It is common to use cars with hundreds of TOPS computing power) .

This is also the core logic that distinguishes the commercial vehicle track from passenger cars - cost control. In order to meet mass production requirements, Inceptio has also formed a "methodology" on its autonomous driving technology stack.

The core technology behind Truck NOA is divided into three major parts: end-to-end network with safety guardrails , “Super Driver” , and Inceptio’s self-developed smart driving platform .

Common autonomous driving solutions are mainly based on modular architecture, usually including perception module, positioning module, decision-making module, path planning module, control module, etc.

However, the disadvantage is that each model requires separate training and optimization iterations, which requires large R&D investment, and errors may be amplified step by step, and repeated calculations are prone to occur between each module, which wastes computing resources.

The end-to-end network began with the publication of the paper "End to End Learning for Self-Driving Cars" in 2016, integrating multiple small models into a "large model", collecting raw data through sensors, and The raw data is input into a unified deep learning neural network (large model) and driving commands are output directly.

However, due to its inexplicability, insufficient reliability, and lack of security guarantees, traditional end-to-end networks are difficult to apply to real scenarios on a large scale.

Inceptio Technology proposed the "end-to-end network with safety guardrails", which decomposes the traditional end-to-end network into several differentiable sub-networks, establishes an explainable end-to-end network, and constructs safety guardrails for constraints.

Yang Ruigang , CTO of Inceptio Technology , introduced that the so-called safety guardrail can be understood as two systems running at the same time, one based on the end-to-end model, which is the main system. Occupancy grid technology (Occupancy Grid Map - OGM) is used to represent in three-dimensional space to achieve effective recognition of special-shaped objects and complex scenes.

In addition, in response to the challenge of long sensing distance of heavy trucks that consumes OGM computing power and memory, Inceptio has developed efficient OGM technology, which uses adaptive granularity and sparse algorithms to reduce computing power consumption by 55% and memory consumption by 70%.

In addition, the safety guardrail has some small modules in the middle of the end-to-end model. The small modules will generate interpretable results. The interpretable results form some regular solutions, which are then weighed and compared with the output results of the main system.

Moreover, Yang Ruigang also emphasized that Inceptio uses real data when training the end-to-end network.

"Super Driver" is a large driving behavior model developed by Inceptio for the special usage scenarios and needs of heavy trucks - TruckGPT .

TruckGPT atomically subdivides driving behavior and scene data in more than 300 dimensions such as weather, road surface material, lighting, road structure, traffic flow, vehicle type, load, etc., and converts it into natural language; at the same time, it combines manual annotation and automatic annotation to form a formal The driving behavior atom set of samples and negative samples is used to fine-tune the LLM model, striving to surpass experienced drivers in terms of fuel consumption, safety and driving comfort.

Finally, Inceptio took the lead in choosing to develop a full-stack self-developed intelligent driving hardware module, which was driven by both mass production cost considerations and the inherent law of integrating software and hardware for autonomous driving.

Inceptio's latest ADCU computing platform addresses the challenges posed by long-distance weak network connections and changing working conditions on trucks in closed-loop data. The new platform provides large-capacity storage capabilities and a data management system, through data compression, intelligent breakpoint resume and other technologies. Ensure that the success rate of high-value data return is 99.9%.

The new ADCU has also reached ASIL-B functional safety level and ISO 21434 certified information security level, fully coping with the harsh operation and maintenance environment of trucks.

The technology behind the launch of the world's first truck NOA highlights three characteristics: first, emphasis on functional safety; second, cost and engineering feasibility; and finally, fuel-saving and smooth features that closely fit the commercial vehicle track.

There is no radical emphasis on fully unmanned or L4, nor is it deliberately showing a "XX mileage does not take over" demo, it is pragmatic and a bit "simple".

But this is exactly the know-how accumulated by Inceptio from mass production of smart trucks, and it is also the core that enables it to quickly launch commercialization and achieve an operating mileage of over 50 million kilometers in more than a year.

How to implement truck NOA?

First, let’s quickly explain the biggest and most essential difference between smart trucks and passenger cars or Robotaxi tracks.

Passenger cars serve C-end users, and the products come in various shapes and sizes, with high and low smart driving solutions and differentiated functional highlights.

However, the core target of commercial vehicle services is the logistics industry, the B-side, and the only concern is cost .

To put it bluntly, commercial vehicles make money by pulling goods. The lower the cost, the greater the value. If you save a little, the fleet or driver can earn more.

The full cycle cost (TCO) of commercial vehicles has become a key indicator, including driver costs, which account for about 1/3; fuel costs, which account for about 1/4-1/5. The rest also includes maintenance, accident insurance, etc.

To give you an intuitive data concept: Generally, the scrappage limit of a heavy truck is about 1 million to 1.5 million kilometers. A fleet or driver who travels frequently can usually reach this mileage in two to three years, and the total cost is about 2.5 to 3.5 million yuan.

Whether it is labor, fuel, maintenance, etc., even if the driver can save 1% of the cost, the driver can earn tens of thousands more a year. The more he saves, the more customers will be willing to pay.

Focusing on this core appeal, Inceptio Executive Vice President Ayushun introduced the current commercialization progress:

The commercial operation mileage exceeds 50 million kilometers, and the number of smart heavy trucks is close to 700. It covers more than 340 trunk high-speed operating lines in seven major core economic zones across the country.

A total of nearly 50,000 shipments have been made, all of which were completed by a single driver (one driver) .

In more than 600 days of operation, autonomous driving mileage accounted for more than 90%. Within the autonomous driving mileage, there will be no accidents and no insurance compensation.

From this description, we can see the main logic behind the implementation of Inceptio smart trucks.

Truck NOA brings fuel savings . It is a direct commercialization advantage.

According to Ayushun, the fuel-saving performance of Inceptio smart heavy-duty trucks is better than the customer's fuel consumption assessment standards, and the average fuel saving per 100 kilometers can reach 1 to 3 liters per 100 kilometers.

Compared with excellent human drivers, fuel savings can reach 3%-7%. Among them, 30% of the normal operation lines can achieve a 7%-10% reduction in fuel consumption.

Secondly, truck NOA allows drivers to "save effort", which is the prerequisite for the second commercial advantage.

Tracking research data by the Beijing Institute of Technology's aerospace human factors engineering team shows that the physical fatigue of smart truck drivers using Inceptio Trucks NOA is about 35% lower than that of drivers using traditional trucks, and the psychological fatigue is about 11% lower.

This means that in long-distance freight transportation scenarios, tasks that used to require two or more drivers can now be completed by one driver.

For example, in various trunk logistics segmentation scenarios such as express delivery, contract logistics, less-than-truckload and vehicle dedicated lines, Inceptio's customers have now achieved a significant reduction in the ratio of people to vehicles, which can save 20% to 50% of labor costs.

This is "saving people".

In addition, another advantage of the launch of Inceptio's smart truck products is the smooth and safe operation of the truck's NOA, which reduces cargo transportation losses and vehicle insurance costs.

Summarizing Inceptio’s business implementation exploration, it shows the following progress:

With a strong technical base and its main competitive advantages of "fuel saving", "two drivers into one" and "safety and cost", it has been recognized by the logistics industry.

And on the sales side, it also covers customers in multiple scenarios such as express delivery, contract logistics, and LTL.

In the entire industry, most smart heavy-duty truck players are still overcoming the problem of mass production of front-end equipment, and some have not even found a suitable landing scenario.

Inceptio’s commercialization is currently the largest and most profound, and it is also the first to initially explore a landing model.

Throughout the entire track, Inceptio Technologies took the lead.

The “scarcity” of Inceptio’s model

What do you think of Inceptio’s truck NOA and commercialization progress?

Due to common conditions, the entire smart truck racing landscape has changed.

There have always been three stories about smart truck entrepreneurship: vehicle building, L4, and Inceptio’s starting from L3 and gradually moving towards L4.

L4 is currently in trouble, especially overseas. The important reasons are that one is that regulations do not allow it to be released, and the other is that it is difficult to achieve mass production of front-end equipment.

In fact, the reason why the "new truck force" was born is because the L4 model mass production was blocked. Players entering the game with high-end autonomous driving technology hope to achieve L4 dimensionality reduction by redesigning the underlying platform.

However, the road to building a car involves qualifications, heavy asset investment, etc. Whether it can be successfully passed remains to be verified.

As for Inceptio, which started L3 mass production by cooperating with car companies, due to its commercial pragmatism and strong technical investment, it first figured out the business model - the logic has been established, customers have recognized it, and the rest is to reduce costs. cost.

This is also the second significance of Inceptio Trucks’ NOA of 50 million kilometers:

The turning point for smart trucks has arrived. For the first time, Inceptio has proven the necessity and commercial value of smart trucks, reinjecting confidence into the entire industry and dispelling doubts from policies, capital, and public opinion.

Of course, it also provides the industry with a reference for the “Inceptio Model”.

Ma Zheren, founder and CEO of Inceptio Technology, believes that the so-called “Inceptio Model” does not have “uniqueness” and can even be said to be a “brand”:

Front-loading mass production + full-stack self-research + in-depth operation.

Front-end mass production refers to in-depth cooperation with car companies to add intelligent driving kits to meet car specifications, and the cost is controllable, providing a foundation for implementation.

Full-stack self-research is reflected in the fact that Inceptio is now completely self-sufficient in terms of AI "productivity tools", and it is also reflected in the use of different solutions on a unified hardware platform to meet the needs of various customers.

In-depth operations enable Inceptio to enter the business closed loop as early as possible, iterate on products and acquire customers. In the medium to long term, we will explore how “cargo robots” will revolutionize transportation modes.

In fact, if you take it apart, every one of them has been emphasized by players before. However, Ma Zheren believes that the deep integration of the three and the persistence of it are rare. This is the true connotation of the Inceptio model. It is also the core reason why commercialization is currently progressing the fastest.

Observed vertically, Inceptio Technology’s business progress also shows trends that deserve more attention:

By adhering to the correct strategy for several years, and in the context of mature technology, market, and industrial chain, we have figured out the prototype of a business model and are far ahead of other players on the track. These factors have brought Inceptio to the beginning of a huge growth cycle.

Just like the ideal in 2022 and BYD in 2021.

From an industry perspective, the commercialization formula pioneered by Inceptio Technology will undoubtedly become the industry's formula and guideline, ushering in a new global industry competition.

On this track, China already has the advantage of becoming a global center for technology and commercialization.

Click here ???? Follow me and remember to star~