Blind spot laser radar illuminates the hidden corners of autonomous driving

Why do people think that blind spot-filling laser radar will be the next rollable field? This starts with the number of laser radars rolled by the OEMs. In order to achieve full coverage of the 360° field of view of the entire vehicle and to achieve the blind spots of the sensor field of view as much as possible, the OEMs have piled up several laser radars on some flagship models. Xiaopeng G9 has 2, the Polar Fox Alpha S Huawei HI version uses 3, and the Great Wall Mecha Dragon has 4. The model with 4 laser radars can completely cover the front, rear, left and right directions of the vehicle, while the arrangement of 3 laser radars abandons the rear direction. The models with 1 or 2 laser radars have the radar arranged in the front direction, but the field of view of the left and right sides of the front of the 2 laser radar models will be larger than that of 1 laser radar.

From this point of view, the more laser radars, the better, but laser radars are expensive. From the perspective of the hardware cost of the self-driving system, in addition to domain control, the second most expensive is laser radar. Moreover, the increase of LiDAR will greatly increase the amount of perception information processed by the domain controller, and the chip cost and the cost of flash memory (UFS) will also increase significantly, which will eventually reflect on the vehicle price and lead to a decline in product competitiveness.

Secondly, the technical specifications of the main LiDAR focus on long distances, and the vertical field of view is only 25°~30°, which is different from the large field of view desired for blind spot compensation, and it is not easy to use. Therefore, when pre-researching and planning the next generation of autonomous driving hardware solutions, OEMs and suppliers will seriously consider the product positioning of blind spot compensation LiDAR. The blind spot compensation LiDAR with lower cost but good performance has become a good choice. The blind spot compensation LiDAR is mainly arranged in the lateral position, with a small minimum exposed window size, a detection capability of about 100m, a wide field of view of 100°*75°, and a high perception of road signs, railings, and mezzanines in stereo garages, and a low detection of children, pets, cone barrels, zebra crossings, etc. As shown in Table 1 below, these are the parameters of the mainstream blind spot compensation LiDARs on the market.

Table 1 Comparison of parameters of mainstream blind spot laser radars

Below, I will take RoboSense E1 as an example to elaborate on the role of blind spot laser radar and establish a basic concept of blind spot laser radar. RoboSense E1 uses 940nm laser band, and the laser human eye safety level is the highest Class 1 Eye Safety. It is a safe laser and will not harm the health of pedestrians under normal use conditions. The field of view is 120°×90°, the angular resolution is H×V: 0.625°×0.625°, and the frame rate is configurable 10~30Hz.

Under 100klux illumination and 10% reflectivity, its detection distance is 30m, the detection blind area is less than 0.1m, and the ranging accuracy is ±5cm (1 sigma). The power consumption is less than 15W, the weight is as low as 300g, the operating temperature range is (-40°C, 85°C), and the protection level is IP67, IP6K9K. As can be seen from Table 1, the horizontal field of view of blind spot laser radar is very large, which can support the optimal deployment plan. The so-called optimal deployment plan is to achieve 360-degree coverage of the vehicle without blind spots. As shown in Figure 2, the three deployment plans of 2 blind spots and 1 forward can guarantee the optimal deployment plan, and 3 blind spots and 1 forward and 4 blind spots and 1 forward can guarantee sufficient perception redundancy. This optimal deployment plan can ensure the integration of zero blind spots, detect oncoming vehicles in advance, and make safe U-turns.

Figure 2 LiDAR deployment plan From Table 1, we can see that the vertical field of view of blind spot laser radar is larger than that of forward laser radar. What are the benefits of a large vertical field of view? From the comparison of the 75° field of view and the 90° field of view in Figure 3, we can see that a large vertical field of view can take into account both ground blind spot filling and lateral wide-range perception, effectively reducing the blind spot angle and blind spot range. From Figure 4, we can see that blind spot laser radar can effectively make up for the corner case in the perception of existing autonomous driving hardware.

Figure 3 75° vertical field of view vs 90° vertical field of view

Figure 4: Object perception by blind-complementary laser radar. The author elaborates on the improvement of blind-complementary laser radar on autonomous driving functions from three aspects: parking, high speed and low speed. From the perspective of valet parking, the addition of blind-complementary laser radar can not only perfectly achieve 360-degree coverage of the vehicle without blind spots, but also accurately and reliably perceive the collision boundary, thus realizing a truly safe, intelligent and reliable valet parking function.

From the perspective of highway autonomous driving functions, the current function cannot cope with the cut-in/cut-off scenarios in congested environments. When the vehicle behind rushes in or there is insufficient lateral space, the vehicle cannot change lanes safely. After installing blind-complementary laser radar, the vehicle can effectively increase the lateral perception performance, thereby optimizing the lane change algorithm and realizing lane change in a smaller space. In

low-speed scenarios, mainly traffic congestion functions, the 360-degree full field of view coverage of two blind-complementary laser radars can solve the problems of lateral blind spots and special-shaped target detection, make up for the difference in perception effects in areas outside the field of view of the forward laser radar, and can effectively deal with jamming and lane changes, and effectively deal with complex urban working conditions. Especially in scenes such as turning at intersections in urban areas and mixed traffic of people and vehicles, blind spot lidar can play a more significant role due to the complex composition of traffic participants.

In general, the deployment of blind spot lidar can provide richer point cloud data, enhance the automatic driving system to accurately classify and identify surrounding targets, and help the automatic driving system better understand and predict the behavior and intentions of other traffic participants, so as to timely identify potential dangers and risks, and take appropriate measures to avoid collisions or accidents, thus increasing the practicality of the automatic driving function.