Chen Guanling, technical partner of Fuyou Trucks: Application of autonomous driving in trunk logistics

Guest: Chen Guanling

Organization: Qianshan

Autonomous driving technology is an important part of promoting the intelligent upgrading of transportation facilities. In the complicated transportation system, the road scenes of trunk logistics are relatively standardized. Many experts believe that trunk logistics is expected to become a pilot trial field for the implementation of autonomous driving.

Recently, at the ​​AISummit Global Artificial Intelligence Technology Conference hosted by 51CTO​, Chen Guanling, technical partner of Fuyou Trucks, gave a keynote speech "Application of Autonomous Driving in Trunk Logistics" 》, shared the application and thinking of autonomous driving technology in trunk logistics scenarios from an operational perspective.

The content of the speech is now organized as follows, hoping to inspire you.

In March and April this year, due to the outbreak of the epidemic, local governments upgraded their control and daily supplies were in short supply. Truck drivers rushed to various places to deliver supplies. During this period, extensive media reports also made netizens gradually understand and pay attention to the road freight industry and the truck driver group.

Application of autonomous driving in the field of highway freight transportation

Highway freight transportation is the "main artery" of social operation and the "barometer" of the national economy. In 2021, China's road freight market size is approximately 6.8 trillion, and there are nearly 20 million truck drivers across the country. Moreover, there is a strong positive correlation between road freight and GDP, and the correlation coefficient between the number of active trucks and GDP is as high as 0.86.

But road freight also faces some problems. According to data from the Ministry of Public Security in 2016, trucks accounted for only 12% of all motor vehicles, but trucks accounted for more than 30% of traffic accidents, and 48% of traffic accidents involving fatalities. At the same time, due to falling freight rates, it is becoming increasingly difficult for drivers to earn money, and fewer and fewer young people are willing to engage in this industry.

Spurred by the pain points of frequent accidents and difficulty finding drivers, the application of autonomous driving in the field of highway freight has become the consensus of the entire industry.

(1)Application of autonomous driving in port logistics. The port is a relatively closed and low-speed driving application scenario, which is more friendly to autonomous driving. However, at present, the vast majority of container trucks (referred to as "container trucks") in ports are still driven manually, and the penetration rate of automatic driving of container trucks in ports is less than 2%. It is expected that by 2025, the penetration rate of L4 autonomous driving for container trucks in China's ports will exceed 20%, and the application scale will reach 6,000 to 7,000 vehicles. It is estimated from this that the overall market size of China's port autonomous driving will exceed 6 billion, accounting for approximately 30% of the global market.

There are two reasons why port logistics can become an important implementation scenario for self-driving trucks: First, the implementation of self-driving in ports is fast and the business model is clear. It is expected that in the future Large-scale commercialization can be achieved in 2 to 3 years; secondly, after the implementation of port autonomous driving, it can be extended from point to point, and it can be successfully extended to trunk logistics in 1 to 2 years.

Currently, 13 domestic ports have implemented autonomous driving trucks. Companies including TuSimple, Mainline Technology, and Sinian Smart Driving have begun to promote the commercialization of autonomous driving in ports.

(2)

Highway scenarios are more complex than port logistics, but compared to urban scenarios on open roads, trunk logistics is still more friendly to autonomous driving. In 2021, the number of heavy-duty trucks for trunk logistics in China will be approximately 3.14 million, and the potential replacement market for autonomous driving will exceed RMB 760 billion. By 2025, the potential replacement market is expected to exceed RMB 1 trillion. It can be seen that the market for trunk high-speed logistics far exceeds that of port logistics. Industry leaders in the field of autonomous driving have already made plans for this and are steadily promoting the commercialization of autonomous trucks. TuSimple completed its first fully unmanned test on an open road at the end of last year and plans to implement normal operations. In early June this year, Google’s self-driving company Waymo announced a long-term strategic cooperation with Uber’s freight division on self-driving trucks.

Benefits of autonomous driving to trunk logistics

For logistics companies, in addition to safety, the core driving force of applying autonomous driving is to reduce costs.

In the cost structure of road freight, driver salary costs account for about 25%. The simplest way is to completely reduce driver costs after L4 autonomous driving matures, that is, the overall freight rate will be reduced by 25%. This is a very considerable cost reduction ratio. Of course, we know that the implementation of L4 will take time, and we need to wait more patiently for the maturity of technology and policies.

In addition, generally speaking, drivers will feel very tired after driving 600 kilometers continuously. However, freight orders exceeding 800 kilometers currently require two drivers to complete the work in shifts, that is, one person stops the car without stopping. We predict that the L3-capable autonomous driving system can greatly reduce driver fatigue, and only one driver is needed to complete an order of 800 to 1,200 kilometers. In this way, for orders within these distances, dual driving can be changed into single driving, which can save about 12% of the cost. This is actually a very effective cost reduction measure.

2, Reduce fuel costs.

In the cost structure of freight, fuel costs account for about 23%. Due to the recent continuous rise in oil prices, the proportion of fuel is also getting higher and higher. Reducing fuel consumption generally involves continuously adjusting the engine's throttle to allow the vehicle to achieve reasonable fuel injection. Usually when we talk about the reduction ratio of fuel consumption, it refers to the reduction ratio of the fuel consumption of an autonomous vehicle relative to that of a novice driver.

Specifically, the methods for autonomous driving to reduce fuel consumption refer to:

On the one hand, use high-precision maps and on-board sensing equipment to obtain the road conditions ahead in advance, such as the proportion of uphill and downhill roads, the The speed of the driving vehicle, etc., to make better planning and decision-making - braking and accelerator more accurately.

On the other hand, by accumulating data and optimizing algorithms, the vehicle can be driven in the best state, that is, the vehicle can be kept in the best power consumption range. We have noticed that drivers with good driving habits can reduce fuel consumption by 20% compared to novice drivers, but this requires the driver to be very familiar with the road conditions and know where to brake and where to accelerate. On this basis, autonomous driving can further save 6% to 10% in fuel consumption compared with experienced drivers. Converted to the freight rate, the cost ratio that can be reduced is about 1.5% to 2.5%. In fact, the gross profit margin of many logistics companies is usually only 3 to 4 points, so a decrease in the ratio of 1.5 is actually an effective means to increase gross profit.

Technical challenges of autonomous driving in trunk logistics

1. Perception distance.

Usually the sensing distance of a camera is 200 meters, and the sensing distance of lidar is shorter, about 100 meters. Waymo previously disclosed that their vision system can detect and track objects 300 meters away. The 300-meter distance can ensure that a two-ton car has enough time to stop safely when driving at high speeds.

But for trucks, especially heavy trucks traveling on highways, the weight of the front of the truck alone may reach 9 tons, and the weight of the transported goods can reach 27 tons, so the longer the perceived distance, the greater the weight of the truck. With a longer braking distance, heavy trucks can stop safely.

Of course, not only does the sensing distance need to be long, but the recognition accuracy also needs to be maintained. Because using a telephoto lens to perceive distant objects will result in a decrease in resolution. We have yet to see autonomous driving companies disclose relevant data.

2. Difficulty of changing lanes.

In a high-speed scenario, it takes about 10 seconds for a truck to complete a lane change. If the driver's advance observation is included, it may take longer, and the risk to the safe driving of surrounding vehicles will be greater. big. In most cases, we want to avoid lane changes and avoid safety hazards through detection.

(1) Safe lateral movement. When the car running parallel to you does not want to steal the lane, but the distance between it and your car is too narrow, your car actually only needs to move slightly to the side, and does not need to completely change lanes.

(2) Suddenly grab the road. When a car suddenly grabs the lane, if the decision-making system determines that it is enough to brake and slow down without causing a crash, a complete lane change can actually be avoided.

(3) Actively change lanes. When there are other vehicles merging from the entrance ramp of the highway, if the decision-making system determines that a collision cannot be avoided even with sudden braking, it is a better way to actively change lanes.

All these operations actually require relatively accurate identification of the intentions of other vehicles. But generally speaking, the size, weight, and uneven positioning of the cargo in the trailer behind a heavy truck may cause the vehicle to shift when the driver is driving. Once a deviation occurs, it is easy to cause lateral control instability, which is a prominent control difficulty in trunk logistics driving. Therefore, the requirements for the automatic driving flight control system are also particularly high.

3, Data accumulation.

There is a popular saying that if autonomous driving is to catch up with the safety level of human driving, it will need more than 10 billion miles of road testing. Currently, the routes of autonomous driving in trunk logistics will not cover the whole country. As a test phase, it may only run back and forth on a single route. The purpose is to accumulate data and confirm the algorithm, so the data accumulation process must be very slow.

On the one hand, different lines have different road information, and it is impossible to simply translate the algorithm model of existing lines. On the other hand, from the perspective of development strategy, when the scale of vehicles has not yet begun, autonomous driving companies, as operators, cannot operate multiple lines in parallel. Instead, they will develop new developments after the algorithm model of one line is fully operational. Algorithmic model of another line.

For data accumulation, one of the effective means is to use simulation technology. But if it is just a simple speculation, the simulation mileage does not mean much. For example, some companies will say that our 1,000 kilometers of simulation mileage is equivalent to 1 kilometer of actual road test mileage. This does not actually solve the essential problem. The core still depends on how to integrate important events into the simulation system for testing and Iterative Algorithms. In short, actual drive testing and simulation systems are two essential aspects to improve algorithm accuracy and system security.

Business model of autonomous driving in trunk logistics

Let’s take a look at several different business models of autonomous driving in trunk logistics.

Business model 1:: Provide technical solutions. Autonomous driving companies provide autonomous driving system-related technologies and technical services to OEMs, including sensor configuration solutions, development and iteration of computing platform algorithms, etc.

Business Model 2: Providing autonomous driving operation services, which is a SaaS model. Customer-facing logistics companies. Logistics companies purchase OEM vehicles that autonomous driving companies cooperate with, and at the same time, autonomous driving companies provide autonomous driving technology operation services to them, while logistics companies only need to be responsible for managing and operating the fleet.

Business model 3:: Providing third-party transportation services, which belongs to the TaaS model, that is, Transportation as a

Service. Under this model, autonomous driving companies must build and operate their own fleets and be responsible for the development and iteration of autonomous driving technology.

Business Model 4:: Provide an end-to-end full-process model of vehicle manufacturing, autonomous driving systems and transportation capacity. On the one hand, autonomous driving companies must provide a complete set of autonomous driving systems and third-party transportation services. On the other hand, they must increase their capacity for mass production and delivery and solve the problem of insufficient transportation capacity by increasing vehicle manufacturing.

Currently, it is difficult to say which of these four modes is better, and everyone is in the process of exploring it. However, we believe that autonomous driving companies must go deep into business scenarios and participate in trunk logistics operations in order to better accumulate data and iterate algorithms so that the final value delivered is more in line with the needs of operators.

Fuyou Truck’s approach to facing the pain points of trunk logistics operations

Autonomous driving is a part of trunk logistics operations. In the end-to-end operation process of trunk logistics, there are still a lot of pain points: for cargo owners, it is inefficient to find a car, opaque prices, delayed delivery, difficult tracking of goods, and irregular settlement; for transport capacity, transportation efficiency It is difficult to improve, there is a lack of standardized service capabilities, there is no guarantee of payment collection, etc.

Fuyou Trucks, as an end-to-end vehicle transportation operation platform, uses three intelligent systems (intelligent pricing, intelligent dispatching and intelligent services) to deal with these industry pain points.

#When the cargo owner inquires about the price, the platform will automatically quote the price based on the algorithm. If the cargo owner feels that the price is acceptable, he can place an order. After placing an order, Fuyou uses the intelligent dispatch system to select the appropriate driver to pick up the order. During the entire process, Fuyou uses an intelligent service system to monitor whether there are any abnormalities throughout the transportation process and whether the entire end-to-end process is completely online.

Based on six years of operation results, the on-time transportation rate of our platform has reached 95.2%. In the traditional operating model without technical support, the on-time rate is generally only 80%-85%. In addition, the accident rate is only 2 in 10,000. The empty driving rate is 6%, which can be said to have greatly improved efficiency compared to the industry's empty driving rate of 49%.

Our vision is to move from dispatching trucks driven by humans today, to dispatching smart vehicles that combine humans and machines in the near future, to dispatching fully driverless trucks in the future, to become a true cross-city trunk logistics system. Intelligent operation platform.

The "Venus" plan being implemented by Fuyou Trucks is an open source commercial operation scenario for autonomous driving companies. Self-driving companies are usually companies with a technical background. Their lack of experience in logistics operations often results in low operational efficiency, which also causes them to spend a lot of resources and energy on real operations rather than technology.

We welcome self-driving truck companies to join Fuyou Truck’s commercial operation scenarios. Under the "Venus" model, Fuyou Trucks is responsible for the distribution of goods sources, driver management and delivery quality. Autonomous driving companies only need to focus on improving technology.

On the operation platform, we will fairly monitor some key technical indicators, such as average takeover mileage, fuel consumption data in autonomous driving mode, emergency braking and emergency stopping. Speed ​​and so on. At the same time, Fuyou can also endorse these self-driving truck companies with a certain amount of income. Our freight rates can be passed on to self-driving trucking companies. In addition, we can also give priority to purchasing some trucks with relatively mature technical solutions.

For more exciting content, please see the official website of the conference:​​Click to view​​

The above is the detailed content of Chen Guanling, technical partner of Fuyou Trucks: Application of autonomous driving in trunk logistics. For more information, please follow other related articles on the PHP Chinese website!