Author: Yan know wisdom car annual meeting

At the “First Yan Know Car Annual Meeting”, Lu Haifeng, Vice President of Shanghai Daisi Intelligent Technology Co., Ltd., shared his application practice of the P-Box, accelerated the progress of automatic driving technology and built components for the theme of “P-Box in 2021”.

Automatic Driving Without High-Precision Maps Is Painful

After the COVID-19 pandemic, L2 has become a trend, with cars like Corolla and Feidu also having L2, and L3 has become a major trend.

From level 1 to level 5 of automatic driving, L3 or higher was once considered an unrealistic concept. However, experts from various fields in the industry have a lot of expectations for various Pilot functions today. The so-called “hands-free” is very attractive in both high-speed and congested road conditions.

Why use a “positioning box”? Instead, why do we need high-precision maps? The positioning box is a supporting component for high-precision map services. Personally, driving is a physical labor that is not very enjoyable for me. Especially sometimes driving is painful. I am not familiar with the road conditions and often get on the wrong ramp. Without high-precision maps, automatic driving is also very painful.

I have experienced many advanced assisted driving vehicles without high-precision maps, and when I found that this road couldn’t go straight, I had to make a hard lane change. This made passengers feel sudden and lost their sense of security.

Therefore, the “electronic horizon” is a popular concept now, and it has been proposed for many years. At least 3 kilometers forward and 1 kilometer backward, this is the biggest possibility that automatic driving brings us.

Among the three major sensors, only one-camera is currently used for positioning, which refers to the lateral lane keeping. As automatic driving develops very rapidly, lidar and mm-wave point cloud will be introduced soon. Until now, the high-precision positioning produced by XPeng has been a triple-redundant system.

GNSS, the Global Navigation Satellite System, INS, the Inertial Navigation System, and CAM, the Camera, all have their own advantages and disadvantages in terms of positioning. The larger the dot, the lower the accuracy, and once it reaches a certain level, it becomes red, which is unacceptable (NO GOOD).

As time goes by, the dot will become larger, and the circle will also become larger. The INS has a very high density and refresh rate that GNSS cannot match. Its robustness is complemented by the camera to a certain extent, such as in some backlighting scenes or scenes with changes in brightness, where the camera may have a momentary failure at the millisecond level.

Introduction of P-Box Electrical Architecture

Therefore, we say that many components in a car cannot be said to be better than others, and they all have their own strengths. As the saying goes, three humble craftsmen top a wise Zhuge Liang. Basically, cars are heterogeneous redundancy, complementing each other to achieve complex functions.

The so-called positioning box is a combination of an inertial navigation system, which is now commonly called P-Box by OEMs. Its position in the automatic driving electrical architecture is shown in the figure below. As a small ECU responsible for high-precision positioning, it provides positioning data to the map module in the domain controller to match with high-precision maps.

So, what kind of accuracy does HWP positioning need to be enough? One is inertial calculation, and the other is RTK (Real-time Kinematics).

First of all, assuming that a car is driving on a highway with a national standard lane width of 3.75 meters, we hope that it will not deviate laterally from the lane for approximately 0.87, including straight lines and turns. In this process, under most general working conditions, high-precision satellite corrections, wheel speed, and cameras all work normally and verify each other.

However, in some special working conditions, RTK may not work, and cameras in some places may not work. This is where the P-Box comes into play.

# This is the English version of the Markdown text, which keeps the HTML tags and only output the results in a professional way:

When decomposing the inertial errors, we need to understand why there are errors in its calculation. The main reason is the non-ideality of sensors, such as zero offset and scale factor. Zero offset means that even when there is no turning, there will be a certain angular velocity displayed.

Through simple decomposition, such as letting the error factor be 50% to 50%, each contribution cannot exceed a 0.435-meter error. Using a 3 Sigma chart, such as reaching 99.7% for 3 Sigma, we can see what the indicators can achieve.

Later, a series of formula calculations were carried out. For example, a simple integration formula derivation was performed, and three different correlation scale factor errors were listed: 0.3 ‰, 0.4 ‰, 0.5 ‰.

Let’s take a look at the zero offset part. One is the instability of zero offset + temperature zero offset. Temperature change leads to zero offset, which is a characteristic of MEMS sensors. In the decomposition, such as 50% to 50%, focus first on the instability of zero offset, and the indicators are relatively low, and there is even no data available in the manual. In the past, it was >10 degrees/hour, now the performance is getting better, and the cost is getting better. According to the 50% to 50% algorithm, its gyro zero offset cannot be greater than 3.6 deg/h.

For IMU, the indicators are an endless pursuit, but in engineering, cost-effectiveness often needs to be considered. Many indicators can be improved by 10 times or even 20 times through calibration, which is a cost-effective engineering method. Cost-effective sensors are first used to improve performance through calibration, and drift zero offset and scale factor are good examples.

In short, in the above case analysis, the accuracy of the IMU needs to meet the requirement that the 3 Sigma lateral error is less than 0.87 meters. Shanghai Daishi provides the latest generation of P-Box for front-loading, which specifically provides a front-loading combination inertial navigation system for OEMs. It can be said that after 7 or 8 years of development, there are finally mass-produced components for front-loading. Similar products with similar indicators in the past may have sold for more than tens of thousands of yuan. In 2022, everyone will soon be able to buy cars equipped with this new generation of motion and position sensors on the market.After years of iteration, dual-frequency RTK and single-frequency RTK have become a standard configuration in the industry. Domestic enterprises providing similar high-precision positioning services include Qianxun, and they provide differential correction data to satellite modules, which is a data service then calculated by computing units. An analysis was conducted using RTK positioning data on a highway in a first-tier city in China. It can be seen that the positioning accuracy of single-frequency RTK is far lower than that of dual-frequency RTK in many areas. In fact, BeiDou, GPS, and GLONASS all have two frequency bands, but this requires satellite computing power and the chip power needs to keep up.

Another point is that dual-frequency redundancy leads to much better performance stability. The concept of CEP (Circular Error Probability) here is a military concept, which refers to the circle where 95% of the drop points are located. If it is single-frequency RTK, this circle is larger than 0.4 meters, while dual-frequency can reach 0.2 meters. It can be seen that many cars have been using single-frequency RTK before, which is not a good initial condition for inertial calculation and reduces performance by 46%. Although it has been well-equipped with P-Box and IMU, due to the single/dual-frequency issue, the initial error of single-frequency is presented here and no matter how well it is calculated, it is easy to deviate from the lane.

The development trend of P-Box is as follows: The first generation of P-Box is an ESP-level product that does not require gyro. Some companies produced single-frequency RTK two years ago. The fast-spreading second-generation P-Box has a higher level of IMU, with full temperature range calibration, CANFD, and Ethernet (as Ethernet is relatively expensive, not all companies will use it). The third generation considers functional safety. Relying solely on 4G network differential links cannot solve the problem, so satellite-based differential correction such as SSR is needed, which is a cost issue. Can consumers use it? Maybe an annual fee needs to be charged, similar to Tesla. There is a lot of imagination space now, which may not have been considered before, and selling cars used to be a one-time deal. NovAtel’s products used to be military and civilian grade, costing five figures. Now, the products we produce are in the thousands, with low posture and low cost. If it reaches the third generation, functional safety may cost more than 1000 yuan, but it will not be much worse, and optimization solutions can also be made in subsequent service fees.

From military to civilian.Diasys Intelligence is in its seventh year and has engineers who specialize in automotive electronics, from flight attitude sensors to the automotive electronics field. The company values technology, quality, and production capacity. In 2020, Diasys built its own fully automated production factory in Shanghai with self-developed technology and equipment, holding all intellectual property rights.

The company is dedicated to developing and producing components that accelerate the advancement of autonomous driving technology. Its three major product series include position ECU and IMU for front loading; the mapping-level red box series, which is extremely accurate and has a much higher cost performance than military-grade products; and products targeting traditional testing industries such as chassis testing, parking, and networking.

By 2025, the concept of autonomous driving will have been around for 100 years, and many Chinese companies may emerge as leaders in the industry, as China seeks to become self-sufficient and domestic manufacturers are willing to embrace new technologies. Never before has autonomous driving been so close to becoming a reality.

Diasys has already served nearly 100 companies and, in the future, will use its superior engineering capabilities and technical services to serve a wider range of intelligent-related fields.

This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.