What kind of lidar system is needed for mass production of autopilot?
In 2015, vehicle borne lidar was introduced into the Chinese market. From high-level equipment in autopilot forward-looking research to core components in mass production projects, the industry's understanding of lidar is gradually deepening.
In the past five years, Liangdao intelligence has been involved in the development of the industry. Starting from "lidar promotion" and "perception test verification", the target of lidar vehicle mass production has never changed.
In 2021, lidar will enter the rush year of mass production. The products of Liangdao intelligent are further improved. In this article, we'd like to talk with you about our understanding of "lidar vehicle mass production" and what we're going to do next.
△Photo by S&B Vonlanthen on Unsplash
Understanding of lidar vehicle mass production
In the past year or two, everyone has been paying attention to the latest lidar which is more suitable for vehicle environment. Hardware companies actively develop a new generation of solid-state products to prepare for the hardware test and international standard certification that must be passed before boarding.
But the fact is that lidar is not a plug and play hardware. The main engine factory still needs to care about many problems：How to select a suitable hardware from the lidar samples with different technology paths? How to develop algorithms for hardware, how to test and verify the perception ability to ensure mass production requirements? How to solve the engineering problem of hardware and software integration? These problems are related to mass production.
Therefore, Liangdao intelligent believes that there are still many tasks beyond hardware preparation. You might as well see what work to complete from the boarding process of sensors：
● Therefore, Liangdao intelligent believes that there are still many tasks beyond hardware preparation. You might as well see what work to complete from the boarding process of sensors
There is no lidar that can cope with all automatic driving scenarios. The selection and customized development of new sensors should be jointly completed by the supplier and the main engine factory: define the automatic driving function according to the application scenario, further refine the sensing function, and finally correspond to the lidar sensing coverage, how far to see, how clear to see, and which targets to identify……
Therefore, mass production of the main engine plant and suppliers are not small challenges. The main engine factory should have the ability of function definition. The supplier should not only be familiar with lidar products, but also have rich experience in application development. After many discussions, we can finally determine the hardware performance, hardware deployment mode and software algorithm to achieve the function, and select the products that meet the requirements from the numerous lidar hardware on the market.
Software algorithm development of mass production lidar
After determining the hardware and installation location, the next step is targeted algorithm development. What we are most concerned about is what functions the algorithm can achieve and which objects can be identified and distinguished. But in fact, this is only 5% of the mass production algorithm development work, and the remaining 95% is special algorithm development for various special driving environments and road conditions. For example, in the development of the algorithm, the factors such as whether the vehicle has suspension or not, the front of the vehicle cocks when there are passengers in the rear, etc. should be taken into account.
In addition, automatic driving is a safety related product, and redundancy safety should be fully considered. Therefore, the development of lidar sensing algorithm should also meet the requirements of functional security. In the process of mass production algorithm development, the concept of functional safety is integrated.
Perception test verification
In the complete sensor development V model, the workload of test and verification is almost equal to that of development. This is the work that can't be skipped before lidar gets on the car safely.
The best strategy to verify the sensing ability of sensor is to evaluate its performance in real environment. The lidar to be tested and the truth value sensor should be carried on the test vehicle, and the data should be collected at the same time. Finally, the perception results of the two sensors should be compared, and the truth value should be used as the "reference answer" to evaluate the performance of the lidar. Through SIL and HIL based on real road data, the algorithm is further trained and tested until it meets the requirements of mass production function.
The test and verification of a mass production level project needs to build a vehicle data acquisition platform and data management platform. It takes about 1-2 years to complete the perception test verification. The test team will drive a million kilometers, and the amount of data collected and managed will reach Pb level.
Finally, there is a factor as important as the sensing reliability and detection accuracy of lidar - the cleanliness of lidar. Professionals have pointed out: when the lidar works close to the sensing limit, such as detecting small objects on the highway in the distance, it will try to find the smallest signal. At this time, any dirt that blurs the vision is very destructive.
Therefore, the cleaning system will directly affect the sensitivity of the sensor, and its importance is self-evident. And, just as there is no all-around lidar, there is no cleaning system that can meet all the needs of autonomous lidar. According to the scene where the sensor is used, such as hot, cold and rainfall in the area, the cleaning system needs to be customized.
What does Liangdao intelligence do?
At the beginning of its establishment, Liangdao intelligent focused on the route of automatic driving L2 + to L3 + passenger cars, and entered the market through the direction of lidar intelligent perception software.
On the one hand, we think that software and algorithm development, data mining is the core ability. The modularization and standardization of lidar hardware will become the trend of the times. Each technology route hardware needs to have the ability to "intelligently" sense - which means powerful algorithm and software support, as well as a large number of scene training and testing.
On the other hand, this direction can give full play to our team advantages. The key technical personnel of Liangdao intelligence were responsible for the development of lidar algorithm in European automatic driving mass production project, and accumulated valuable experience.
Therefore, since 2017, we have built an automatic test and verification system for the development of sensing function, realizing the closed loop of "development test" from lidar. Up to now, this capability is relatively scarce at home and even internationally.
In this process, we cooperate with the lidar hardware head enterprises of various technical routes to understand the differences in hardware performance, and do perception development for different hardware to further consolidate the algorithm capability< br />
We are proud that in 2019, we participated in the world's first solid-state lidar vehicle specification mass production project by providing sensing system test and verification services< br />
In 2020, we have witnessed the first blowout period of customers' demand for lidar vehicles, and also found that customers' demand is becoming increasingly clear: < br / >
1. The mass production node is getting closer and closer. Customers need to complete the adaptation, integration and function development of lidar in mass production within a limited time. The customer hopes that the supplier can join in from the function definition stage to achieve the target of laser radar boarding< br />
2. The mass production of autopilot is inseparable from the local development, but there is no precedent of mass production of solid-state lidar autopilot in China. OEM is looking forward to a more open mode of cooperation, and hopes that suppliers with localized service capabilities can grow up quickly and jointly achieve the mass production target.
In the face of changes in customer demand, Liangdao intelligent has been improving the product "add".From 2021, we will further improve the product matrix on the basis of deep ploughing perception system test and verification service, provide lidar system solutions together with lidar hardware and supporting equipment partners, and support customers to complete lidar performance definition, product hardware selection, perception algorithm development and test and verification before mass production, And the mass production target of system integration into the whole vehicle.