Laser radar "getting on the bus" competition, variables emerge

At present, lidar has been regarded by the industry as a "leap" in the logic of high-order intelligent driving and automatic driving. Last week, Tucki P5 of Xpeng Motors, the world's first mass-produced lidar smart car, rolled off the assembly line, and the first owners will deliver it at the end of this month.

Compared with a few years ago when Audi was equipped with Valeo lidar, Tucki P5 may be the first vehicle to really apply lidar to intelligent driving system that can be used on the road.

Why is it possible?

According to the official explanation of Xpeng Motors, the corresponding enhancement function is expected to be realized by OTA in the second quarter of 2022, namely the city NGP function (automatic navigation assisted driving).

In addition, the functions of urban NGP need to be calibrated, tested and verified one by one to improve the reliability. We will start the calibration test in first-tier cities (except Beijing) and gradually expand to other cities, which may last for a long time.

Moreover, the city NGP function depends on the administrative examination and approval of the advanced driving assistance map of each city (some administrative districts may take longer), and this function cannot cover all cities when it is pushed, and will be gradually opened according to the examination and approval of each city.

According to the plans of other car companies, the Great Wall Mocha version equipped with lidar will be listed before the end of the year. Next year, Weilai ET7, Volvo XC90, Zhiji L7 and other models will also be listed one after another.

Obviously, it is still unknown whether lidar can get on the bus and be really used in mass production system. However, this will not affect the "hardware" of the lidar, because the software can be updated through OTA.

One,

As one of the main sensor options for autonomous driving, the cost of lidar has always been one of the biggest obstacles to large-scale boarding. Although almost all L4 self-driving vehicles are equipped with different types of lidar, even Waymo has only a fleet of 1000 vehicles.

For the passenger car market of private users, the cost sensitivity is higher. Below 1000 yuan, it is the baseline for mass production.

At present, most laser radar suppliers are still "struggling" to reduce the cost from 1000 to 500. For the goal of $300 or even $ 100, some suppliers choose the strategy of "reducing performance".

What is the reality?

When you see a large number of point cloud images generated by lidar, you can see many details. Such as images of pedestrians, cars, bus stops and even vehicles, but this is not the "real" effect of point cloud, because most of the data will be discarded through point cloud processing, which is similar to the traditional millimeter wave radar.

At the same time, the real point cloud object is also "fuzzy", even on the plane like buildings and logo, there will be small bumps (not real planes), which means that the returned points will rebound on the subtle changes of surface texture, or because the algorithm used to remove the vehicle's own motion will produce small errors.

Even a stationary object, its position will change slightly. Because the lidar only returns a series of points on the surface of the object, it is necessary to filter out the jitter by Kalman filtering and other algorithms.

But it is a double-edged sword for algorithm processing.

Because if there are too many filters, the perception system may "miss" the actual moving object; However, if the filtering is not enough, the system may calculate that an object has just moved a perceptible distance in milliseconds, which may lead to misjudgment of subsequent path planning.

When some companies claim that their lidar "sees" 200 meters or 250 meters, what they really mean is that the sensor is sensitive enough to detect the pulse returning from a certain part of an object at a certain distance, but this is almost meaningless to the system.

This means that the point cloud density is very important.

Because the working principle of lidar is that the density of point echo decreases linearly with the function of sensor distance in both horizontal and vertical fields of view. When the density of point cloud is low, it will seriously affect the object segmentation and classification, and it becomes difficult or even impossible to determine the object type.

For example, DJI Livox's non-repetitive scanning mode and unique flower scanning mode, with the increase of fusion time, the field coverage of point cloud will increase continuously until the field coverage is close to 100%. Under the traditional mechanical scanning, when the lines are not dense enough, there is the possibility of losing objects.

Previously, Livox's horizon series had a detection range of 260m and a horizontal visual field of 8 1.7 degrees, covering four lanes with a range of10m. However, in the middle and high speed scene, this method has the delay problem of accurately identifying obstacles.

On the customized version of Livox HAP provided to Xpeng Motors, Livox added two scans of ROI area, thus enhancing the safety perception of pedestrians and cyclists.

For long-distance detection (more than 300 meters), Livox has also introduced a new solution called Avia, which can switch between different scanning modes, ranges and different scenes. Among them, the repeated scanning mode is used to meet the application requirements of high precision and dense point clouds in specific areas.

The purpose of this improved method is clear, that is, to increase the point cloud density in some areas.

From the technical dimension, ranging, points per second (PPS) and angular resolution in a given field of view (and the corresponding field of view angle) are three main specifications, among which the third parameter is used to determine the difference of target detection and classification capabilities.

At the same image refresh rate, a higher PPS multiple means that the number of points on a given target increases significantly at the same distance (R), which directly improves the higher accuracy of target detection and classification.

In Xie's view, vice president of Innovusion's autonomous driving sales market, why emphasize that lidar should look far, not at vehicles with a distance of 250 meters, but at small objects with a distance of 100- 150 meters, so that the system can have a safe early warning time and distance.

This requires point cloud density, especially for objects with low reflectivity.

For example, 100 photons are emitted, and only 10 photons are returned. Only when vehicles with reflectivity of 10% can be seen at 250 meters, it is possible to see roadside tires with low reflectivity of 150 meters, and it is possible to see small black objects (with lower reflectivity) at a closer distance of 100 meters.

Behind this, thanks to Innovusion lidar's long-distance detection and high-definition resolution, pedestrians beyond 120 meters can get more than 20 points or vehicles around 400 meters can get more than 20 points, which is very important for the subsequent perceptual recognition algorithm.

In addition, Innovusion lidar also has the function of dynamic focusing. Through local pixel encryption, it can "stare" at key targets and small objects in the region of interest and obtain more accurate three-dimensional information.

Second,

At present, one of the cost reduction strategies of lidar is to combine lidar with different performance specifications. However, the problem is that the system may need more lidar to provide full field coverage, or it may be supplemented by other sensors. This means more requirements for data processing, sensor fusion, high data transmission and high computing power.

Taking the scheme of Ibeo as an example, the performance parameter of ibeoNEXT is to detect the 260-meter target at the horizontal field of view angle of 1 1.2 degrees, and the field of view angle of 32 degrees is still under development. This means that two laser radars with blind compensation are needed to form a wide coverage environment perception of the road ahead.

For example, on the basis of self-developed and mass-produced short-range flash lidar, Continental plans to start producing long-range lidar from 2024 by participating in the lidar company AEye last year.

On the other hand, Yijing Technology has launched a complete set of MEMS lidar solutions, including large field of view MEMS lidar for short-distance applications and forward long-distance MEMS lidar based on 1550nm fiber laser.

Undeniably, the large-scale application of lidar needs to greatly reduce costs, improve product life, and break through technical bottlenecks such as higher detection distance (more than 250m, even 300-400m) and ultra-high scanning beam.

1550nm is one of the mainstream choices. No matter Yijing Technology, AEye or luminar, this scheme is adopted to design long-distance detection lidar. Using 1550nm laser can not only improve the optical output power by several orders of magnitude within the range of human eye safety, but also effectively avoid the sunshine noise area, thus reducing the background light noise.

It can also be seen from the 1550nm+MEMS lidar ML-Xs introduced by Yijing Technology that all the parameters have reached a new height. For example, the field of view angle reaches120 25, the angular resolution reaches 0. 15, the wire harness reaches 200 lines, and the background light noise (under natural lighting conditions) is reduced by 70%.

In addition, 1550nm transmitter is safer than 905nm transmitter, which can increase laser power, improve signal-to-noise ratio, reduce pulse width, be safer for human eyes, and more importantly, improve the effective distance of laser radar.

But at present, the biggest obstacle of 1550nm scheme is cost. The vertical integration of core supply chain system is also the key link to reduce costs and ensure upstream supply in the future.

Luminar acquired InGaAs chip company OptoGration Inc and chip design company Black Forest Engineering. The main layout is 1550nm InGaAs photodetector chip and special data processing chip. Ideally, the cost of mass production can be reduced to a few dollars.

Yijing Technology also selects the underlying chips and components for independent research and development and innovative design. At present, it has formed its own proprietary lidar chip and core algorithm, thus further reducing the cost of 1550nm lidar.

In the eyes of the industry, "the core electronic components of lidar are being integrated into ASIC, which has the advantages of higher density, lower cost and higher reliability. This trend roughly follows Moore's law of integrated circuits, which means that it is possible to greatly reduce the volume, weight and cost of lidar. "

The 1550nm scheme introduced by AEye emphasizes that it can be placed behind the windshield, similar to the traditional front camera. This is very important for future vehicle design, which will not affect the appearance, but also reduce the constraint of wind resistance coefficient that may be caused by external installation.

However, in terms of cost and supply chain maturity, 905nm still has specific advantages, although this wavelength brings concerns about eye safety (such as increasing power) and limits the detection range.

In addition, from R&D to manufacturing (product yield indirectly affects cost), supply capacity and after-sales support, lidar suppliers need to prove to the market that sustainable and efficient mass production is ready.

Before that, there were many variables in this market.

Third,

Among many variables, one is very critical, and that is the standard.

As we all know, in the automobile pre-assembly market, in addition to the default certification achieved by a series of industries such as vehicle classification and functional safety, there are also relevant performance requirements and testing standards in different regions and markets. For brand-new automotive electronic components, lidar is no exception.

According to the data of Advanced Engineering Intelligent Vehicle Research Institute, with the rapid growth of L2 new cars in China in 2022-2023, the lidar for advanced intelligent driving will enter the first round of growth cycle. It is estimated that by 2023, the scale of passenger car lidar in China will exceed 6.5438+0.5 million.

This means that it is very clear how the technical route will develop and what choices the market will make: the mass production of lidar has begun. The whole industry also urgently needs standardization, thus providing reference for large-scale pre-assembly mass production.

10 12 On June 2nd, the Electronic and Electromagnetic Compatibility Sub-committee of the National Automobile Standardization Technical Committee organized the inaugural meeting of the drafting group of "Performance Requirements and Test Methods for Vehicle-mounted Lidar", which initially established the standard system composition of vehicle-mounted lidar and the plan and division of labor for standard formulation.

Among them, in terms of national standards, Jose is the lead unit, Baidu is the joint lead unit, and * * * is also responsible for formulating the national standard GB/T "Performance Requirements and Test Methods for Vehicle Lidar". In addition, Jose also took the lead in formulating some laser radar industry standards.

Just in September this year, Panda128 Lidar of Wosai obtained the world's first functional safety product certification of laser radar ASIL ISO 26262 B issued by SGS. As a globally recognized automotive functional safety standard, ISO 26262 is one of the main entry barriers for the pre-positioning of core components in the field of intelligent driving.

For OEMs and first-class parts suppliers, the drafting of GB/T "Performance Requirements and Test Methods for Vehicle-mounted Lidar" means that all kinds of products on the market in the future can be compared under the same standards, which reduces the threshold of OEM selection and additional hidden costs.

Judging from the minutes letter of the standards drafting meeting of the Automobile Standards Committee, a preliminary "opinion" has been formed on the improvement of the standards. More practical functions such as self-check, fault alarm, start-up time and wake-up function are added to the basic framework, and the point cloud performance requirements of high-speed objects and high-dynamic scenes are further distinguished in combination with actual scenes.

Some suggestions are put forward to further standardize the test conditions of lidar, such as laboratory light source and reflectivity of test board. The test layout is described in detail. The IEC standard reference provided by Rheinland is given for the safety of human eyes. It also gives comprehensive requirements for environmental testing of vehicle regulations.

At the same time, in the global market, IEEE also started to formulate standards for performance testing methods of lidar last year, focusing on performance testing methods, including distance accuracy/precision/resolution, maximum/minimum distance, detection probability, angle accuracy/resolution, reflectivity and so on.

At present, the ambiguity of the requirements of the functional scope of lidar has made clear the urgency of the relevant standard test methods, which means that neither lidar suppliers nor automobile manufacturers can make a horizontal "comparison" of products, and the industry lacks transparency.

In March of this year, Luminar reached an agreement with Zenseact, a software subsidiary of Volvo Cars, to integrate lidar hardware and sensing software to provide a complete set of autonomous driving solutions.

Because, for the OEM, after selecting the laser radar hardware that meets the requirements for the vehicle model, there are still some problems to be solved, such as how to develop algorithms for the hardware and how to test and verify the perception ability of the system to ensure the mass production demand. In laser radar mass production project, software and hardware are equally important.

For example, the sensing algorithm of lidar includes target detection, target tracking, target classification, speed judgment, driving area judgment and even path planning. On this basis, we can realize the development of specific functions. A typical case is that the original point cloud data output by lidar also has corner points.

This means that the lidar sensing algorithm, like the automatic driving system algorithm, also needs a lot of scene data to feed in, and the iteration of the algorithm is realized through various scene verification tests to ensure the safety, reliability, detection rate and accuracy of the lidar system.

Previously, based on several years' experience in closed-loop mass production of lidar software "development+test and verification", Liangdao Intelligent launched a solution for vehicle mass production, which supported customers to complete the definition of lidar performance, product hardware selection, perceptual algorithm development and test and verification before mass production, and the mass production goal of integrating the system into vehicles.

"Scenario-based development and test verification have become an industry trend. The automotive industry urgently needs to introduce technologies such as intelligent sensor software algorithm development, high-quality data acquisition and efficient processing to cope with intelligent changes. " Liang Dao intelligent CEO Xi Ming said.