Have you ever driven a car on the road and suddenly there is a blockage, thinking “there must be road construction, an accident or a traffic police to check the car ahead”, but when you cross the long wait and arrive at the other end, you find that the road is calm and there is no accident what happened?
This is the famous “ghost traffic jam” phenomenon – just because some drivers’ tiny “bad habits”, such as sudden braking, unnecessary frequent lane changes, etc., slow down the car behind, and gradually like a wave wave after wave Passed backwards, causing “inexplicable” road congestion.
At the 2022 WAIC World Artificial Intelligence Conference, SenseTime released its V2X-C cloud control platform. One of the things it hopes to solve is the long-standing problem of “ghost traffic jams”.
“The ‘ghost traffic jam’ is because some bicycles take the optimal decision for the individual, but the optimal decision of the bicycle is often not the global optimal.”
Wu Wei, vice president of SenseTime’s vehicle and road collaborative production and research, pointed out that this requires the aggregation and analysis of the behavior of bicycles at various intersections from the perspective of the cloud, and the optimal instructions are issued to the corresponding bicycles, and then through the vehicle cloud control, so that the bicycle can obtain the optimal decision under the overall situation.
This is just one of SenseTime’s ambitions in the field of vehicle-road collaboration.
In addition to the cloud control platform, the Jueying Vehicle-Road Collaboration Platform released by SenseTime also includes the car-side car-city network platform, the road-side intelligent perception products, and the cloud-based intelligent computing platform.
In fact, under the “vehicle-road-cloud-map” logic of vehicle-road coordination, many high-end players in the industry have entered the game one after another.
So why did Shangtang Jueying choose to enter in a big way at this time? What unique competitive advantage does it have? Can it really open up the industry bottleneck of the vehicle-road collaborative business model?
Cars, Roads, Clouds: Cars and Cities as Infrastructures
In 2021, the Ministry of Housing and Urban-Rural Development and the Ministry of Industry and Information Technology announced two batches of 16 cities as “dual-smart” pilot cities, juxtaposing intelligent networked vehicles and smart city infrastructure to explore the path of coordinated development of the two.
This is the first time to think about intelligent networked cars in cities. In the past, cities were an important support for cars. Now smart cars are trying to become one of the infrastructures of cities. The two are deeply integrated.
This means that in order to develop a dual-smart city, all parties in the industry not only need to understand the business related to vehicles, but also need to understand the business of the smart city, and have a clear insight into which links can generate more value.
Therefore, from the perspective of policy, it can be found that from the closed test field to the pilot demonstration area, and then to today’s dual-smart city, there is a clear trend of vehicle-road collaboration in the city.
Under such a situation, it is just the right time for SenseTime to enter the field of vehicle-road coordination.
From the perspective of vehicle-road coordination, the problem to be solved at this stage is the blind spot problem of intelligent driving vehicles beyond the line of sight, and the problem of how to coordinate multiple vehicles.
This time, the Jueying Vehicle-Road Collaboration Platform released by SenseTime follows the idea of ”smart car + smart road + collaborative cloud” to try to solve the above problems.
Specifically, at the roadside, Jueying’s computing unit products support the fusion perception technology of single vision, vision + millimeter-wave radar, vision + millimeter-wave radar + lidar.
It can achieve a sensing distance of 300 meters, a sensing range of 8 lanes, and an end-to-end delay of 150ms, etc. It can also enable more than 70 vehicle-road collaborative applications to be sent to vehicles in real time, open the perspective of God, and solve the redundancy of bicycle intelligence blind spot.
However, in order to achieve the global optimization of the entire urban traffic, cloud integration analysis and collaborative control are inseparable.
Therefore, in the cloud, in addition to launching the cloud control platform mentioned at the beginning, Shangtang Jueying also released an intelligent computing platform .
It is mainly used to solve some low-level intersection perception problems that only have cameras and no MEC or other sensing devices.
On the one hand, the intelligent computing platform can receive information from the roadside intelligent computing units at high-level intersections. On the other hand, it can also aggregate and analyze traffic data such as video pictures in the cloud to provide perception supplements for low-level intersections and make cognitive decisions. Big data processing.
At present, the Jueying Intelligent Computing Platform has more than 100 algorithm warehouses, covering the algorithm requirements of vehicle-road collaboration, all objects and all scenes.
In Wu Wei’s view, the intelligent computing platform and cloud control platform are like the left and right brains of human beings: discovering traffic problems, making global optimal decisions and sending them to bicycles, they can escort tens of thousands of vehicles in real time every day. The overall efficiency of traffic is “opened up”.
Another product promoted by Shangtang Jueying in the dual-smart city is the Checheng.com platform on the side of the car.
To put it simply, its entire form is similar to a driving recorder with AI computing power . When installed on the car side, the car can become the “eye of the city” for a smart city, and it can detect road diseases that were previously undetectable by fixed sensing units. The situation is fully covered, and then the cloud control platform and AI algorithm are combined to tap more value of vehicle-mounted perception data, and enable “automobile-city collaboration” to take the first-mover advantage.
In addition, following the Robobus driverless shuttle, SenseTime also launched the RoboSweeper self-driving sweeper at this year’s conference, which can also work with cloud control platforms and intelligent computing platforms .
Wu Wei revealed that SenseTime still hopes to be more pragmatic in the field of self-driving cars, so it chose the direction of unmanned cleaning vehicles, which is more expected to be commercialized quickly. The application scenarios are deeply cultivated.
Positioning the midstream of the industry: software-defined vehicle-road collaboration
The concept of vehicle-road collaboration has been proposed for decades. After tortuous exploration, the industry has gradually reached a pragmatic consensus:
The vehicle-road collaboration technology is not a rapid subversion and revolution of the existing traffic management operation mode, but an immersive supplement and evolution of the data source, control method, and release method of the original system.
In this long process, in order for the vehicle-road collaboration technology to develop continuously, it must become part of the business productivity. Only the person who pays the bill is the one who eats the fruit, so that the business model can be opened up.
At present, the commercial closed loop of vehicle-road collaboration still has three major pain points:
-
There are not enough users and users to share the cost of system construction;
-
It has not been smoothly integrated with the existing transportation business system, and still serves autonomous vehicles in general;
-
There is a certain enthusiasm for car companies to participate, but the exploration of the multi-party vehicle-road collaboration software protocol stack is at an early stage.
Solving the pain points requires the cooperation of a huge vehicle-road collaborative industry chain.
From upstream hardware equipment-related companies, such as millimeter-wave radar, lidar, and domestic chips, to midstream intelligent computing-related companies, such as AI platforms, standardized protocol stacks for vehicle-road collaboration, and downstream operations and services , application-related enterprises are involved.
SenseTime, which started with AI technology, clearly positioned itself in the middle of the industry chain, trying to expand its territory in this field with intelligent computing as the entry point.
For this reason, SenseTime proposed the concept of “software-defined vehicle-road collaboration”, which is actually similar to “software-defined smart cars”, and is also one of SenseTime’s most competitive unique advantages.
In the past, some local governments did not have no intention of building roadside units, smart intersections, etc., but considering the hardware update cycle, they were worried that they would have to dismantle and redeploy a batch of hardware in two or three years. The government is discouraged from investing resources to build hardware in advance .
“Our starting point is to use computing power equipment to give the hardware a bottom line, and continuously update the models and algorithms in the hardware, so that even if it has been put into production and used for several years, it can still be used more and more intelligently, and continue to have rich product functions.” Wu Wei said.
Reducing the old is on the one hand, through software-defined vehicle-road collaboration, the hardware life cycle of new projects can also be extended, functions continue to expand, and model accuracy is continuously improved—the marginal cost of hardware construction is thus diluted, and vehicle-road collaboration can be achieved. The investment cost is also partially recovered.
In addition, through the decoupling of software and hardware, SenseTime is also trying to open up the protocol stack for vehicle-road collaboration and build a software operating system related to the vehicle-road cloud .
For example, before, if the system found congestion ahead and wanted to guide the speed of an intelligent networked car, it needed to send the command to the corresponding MEC through the cloud, and then communicate with the car through the RSU, so as to realize the control of the vehicle. guide.
However, similar scenarios cannot be exhausted. Therefore, Shang Tang believes that it is necessary to summarize the application sources that cannot be enumerated, and abstract the corresponding underlying requirements into the software related to the vehicle, road and cloud, and then pass the algorithm warehouse. In this way, the closed loop of a standardized set of vehicle-road-cloud protocol stacks can be opened.
Of course, this undoubtedly requires more linkage with car companies.
At present, it has been revealed that SenseTime is considering integrating its road and cloud-related patents into an open API interface, and working with car companies to develop a corresponding vehicle-road collaboration software protocol stack, which will be used in front-loading and mass production of vehicle-road collaboration models. Explore .
At the same time, SenseTime has also established in-depth cooperation with domestic and foreign OEMs and designated mass production sites, which is conducive to the large-scale coordination of bicycle intelligence and road-cloud infrastructure, and is also conducive to further reducing the hardware cost of mass-produced bicycles for autonomous driving.
Reducing hardware costs, opening up the CLO cloud software protocol stack, and cooperating with car companies… Behind this is SenseTime’s clear understanding of its own capability boundaries, and its open mind to link with upstream and downstream, and it is also SenseTime’s ultimate vehicle-road collaboration solution. A powerful weapon that is going further and further on the road of standardization, modularization and mass production.
AI technology “eats more than one fish”
It is worth mentioning that there is also Shangtang Jueying’s exploration of the business model of the Checheng.com platform.
In addition to strengthening the smart city business and mining the value of more vehicle perception data, the Jueying Auto City platform can also cooperate with car dealers.
In the past, when a traffic accident such as a collision occurred, car dealers generally obtained clues about the accident vehicle through telephone interviews, and then attracted customers to 4S stores for maintenance. But the problem with this method is that it requires very high real-time requirements for vehicle leads. Often, the first call determines which 4S shop the customer will go to for maintenance, which makes the lead price high.
The way of cooperation between SenseTime and car dealers is that since the cost of the Checheng.com platform equipment is at the thousand yuan level, the car dealers are willing to advance the purchase of the Checheng.com platform equipment, and then pre-install it in the new car. Under the premise of obtaining the permission of the car owner, once the vehicle has an accident, the car dealer can quickly get clues and recover the investment.
“We have already passed this one. This year, tens of thousands of vehicles have been laid. Next year, we hope to expand it to the order of one million.” Wu Wei revealed.
Another business path is TO G orientation.
In fact, some commissions and bureaus, such as road administration and urban management departments, do not have the ability to build mobile sensing equipment, so most of the time they can only find relevant departments to coordinate the use of the camera resources of the public security bureau, but the public security cameras often do not cover the needs of the urban management department. road information.
Under the guidance of demand, government departments began to purchase SenseTime’s Checheng.com platform equipment and installed it on buses to serve as daily inspections.
From then on, Checheng.com equipment began to produce road conditions and urban management data, and distributed and pushed it to various commissions, offices and bureaus in real time, so as to charge a certain amount of data service fees.
Whether it is the Checheng.com platform, the cloud control platform or the intelligent computing platform, it is not difficult to find that the backing of SenseTime’s Jueying vehicle-road collaboration platform solution is inseparable from the support of SenseTime’s powerful AI technology.
As a general-purpose technology for “one fish to eat more”, AI has given SenseTime strong cross-industry capabilities, and also the Jueying Vehicle-Road Collaboration Platform’s ability to more efficiently implement algorithm iteration.
From the algorithm level, SenseTime has more than 2,300 pre-training models and more than 300 state-of-the-art vision algorithms based on the OpenMMlab open source system; the “Scholar” model OpenGVLab has more than 3 billion parameters, covering more than 100,000 visual Labels; all things detection algorithms combined with visual recognition technology, there are more than 17,000 algorithm warehouses.
Not to mention the enabling engine and Asia’s largest AI device in Jueying, which further help its vehicle-road collaboration algorithm model to be quickly iteratively upgraded.
In addition, although it has been relatively low-key in the vehicle-road collaboration sector, the story of SenseTime in the field of smart cars can be traced back to 2016, and it is not a new recruit in the industry.
At that time, Honda planned to find an artificial intelligence partner around the world to jointly develop autonomous driving technology and accelerate the development of smart cars. After many rounds of competition with many plans, we finally chose to sign a long-term cooperation agreement with SenseTime, which was established less than three years at the time.
At the same time, SenseTime has not stopped its pursuit of advanced technology, and has transplanted its previous accumulation in the field of AI to the automotive field.
Recently, SenseTime released its interim results for 2022, among which the automotive performance was outstanding. In the first half of 2022, the revenue of the smart car segment increased significantly by 71%, the number of customers served was 20, a year-on-year increase of 54%, and the revenue per customer increased by 11%. Taking smart car cabin products as an example, its mass production and delivery scale ranks first in the industry, and mass-produced car brands include SAIC, GAC, Dongfeng, BYD, NIO, and Chery.
In addition to taking root in the automotive sector, SenseTime has a smart city business line before. For example, in the field of smart traffic management, SenseTime has already developed a set of traffic signal control systems based on video perception and traffic flow parameters. This further gave SenseTime the confidence to open up the closed loop of the CLUCloud protocol stack and the closed loop of the data.
In this way, the Jueying vehicle-road collaboration business instead connects different projects of SenseTime, so as to achieve smooth collaboration between different sectors.
According to reports, at present, the SenseTime Jueying Vehicle-Road Collaboration Platform is gradually being applied to the management of national-level IoV pilot demonstration areas, closed parks, expressways and urban traffic. It has more than 20 smart traffic detection capabilities, such as:
-
Associated with the smart cockpit, it can provide services such as traffic light signal, red light warning, and green light start reminder.
-
Linked to intelligent driving, it can expand the perception boundary and improve the stability and experience of automatic driving.
-
Linking with smart cities can ease traffic, improve traffic efficiency, and drive GDP growth.
“What we want to do is the standardization, modularization, and mass production of vehicle-road collaboration: on the one hand, we can serve mass-produced vehicles with automatic driving and assisted driving; Pain points.” Shang Tang said.
Like a pot of soup that has been boiled for a long time, the vehicle-road collaboration industry has been simmered for many years by all parties, and the soup cooks have been waiting for the soup to boil for a long time.
Wu Wei told Xinzhijia that the arrival of the milestone commercialization node in the vehicle-road collaboration industry will be accompanied by the following phenomena:
-
The penetration rate of vehicle-road synergy (OBU) front-loading mass production vehicles reached 25%.
-
Roadside Intelligent Cells (MECs) will be installed at at least 100 intersections in each city.
-
There are relatively twin or open virtualized vehicle-road collaboration demonstration areas or test sites.
The road is a long way to go. From a long-term perspective, the evolution of the road is actually a hundred years long.
Subversion is not a linear process, but a recursive and unpredictable one.
We look forward to seeing the outbreak of car-road collaboration under the domestic dual-smart city policy, and we can also see the dawn of the 90,000-mile-higher SenseTime Jueying car-road collaboration platform.
Leifeng Network#Leifeng Network#Leifeng Network
This article is reproduced from: https://www.leiphone.com/category/transportation/LYYhxHTHdSC5MgjT.html
This site is for inclusion only, and the copyright belongs to the original author.