Jidu Auto's JET EE architecture

The concept car at Jidu's press conference yesterday attracted a lot of attention. What is very interesting is the EE architecture of Jidu JET - this system revolves around four parts: intelligent driving, intelligent cockpit, vehicle control and powertrain. It connects the entire vehicle's sensors and high-performance computing platform through a high-bandwidth Gigabit Ethernet loop.

Now is the time when new players are starting to play with cross-domain integration and high-computing power computing platforms.

Figure 1. Jidu’s JET EE architecture

Part 1. Current content

The current new forces, from NIO, Ideal to Xpeng, are all centered around the two basic SoC chip platforms of NVIDIA and Qualcomm, and Jidu is no exception this time.

The automatic assisted driving uses Nvidia's dual Orin X chips and the cockpit uses Qualcomm's 8295 chip. The smart cabin domain controller CDC uses Qualcomm's chip. Last year, Jidu, Baidu and Qualcomm announced a cooperation agreement.

▲Figure 2. Intelligent driving and intelligent cockpit

The new features of the smart cockpit mainly include offline voice, 3D human-machine map, in-vehicle and out-of-car interaction, visual perception (personnel monitoring), voiceprint recognition, lip reading capture, etc., which are more focused on the interaction level. Of course, a large part of the 8295 is to drive the screen that runs through it, add some algorithms for identifying people, and more importantly, create an interface for system access.

I have a question here. Is the so-called out-of-car voice recognition function just to set up multiple sound collecting sensors outside the car body, and then collect the sound in a unified way and hand it over to 8295 for processing, which is similar to making a speaker, and then configuring external lights for interaction (the official term is "human-car-environment" natural communication).

Figure 3. Cockpit control

At present, almost all new car companies have adopted the "vision + lidar" automatic assisted driving solution after adopting NVIDIA's Orin chip. Based on the needs of functional safety and training, the two Orin independent dual systems back up and redundancy each other. The intelligent driving domain controller ACU is not surprising.

Note: This cost went up all of a sudden.

Of course, in order to consider the maintenance of the laser radar, a dual laser radar with a lifting hood is used. In the previous description, this design is quite distinctive. According to the layout interpretation of Liangdao Intelligence some time ago, the current Jidu solution has caused a lot of controversy. The function it achieves is roughly the same as the left, middle and right solutions of ARCFOX, and the requirements for anti-fouling are relatively small.

The front hood is directly exposed to the sun and has a high temperature, which will be tested in terms of heat dissipation, waterproofing, collision repair costs and other issues. The traditional hood is a movable part, which poses risks in use and calibration. Strict online dynamic calibration is required to ensure the reliability of the LiDAR hardware. When installed on the hood, the blind area size of the LiDAR is affected by several factors: the vertical FOV mainly includes the installation height of the LiDAR (Height1), the downward tilt pitch angle that the LiDAR can adjust, and the limitation of the hood cover.

Figure 4. LiDAR configuration

Due to the platform cooperation with Geely, the power system and body system are most likely from the Geely Volvo system. This has not been publicized much, and you can refer to the one below to separate the cockpit and intelligent driving.

Figure 5. VIU architecture

Part 2: Gigabit Ethernet and TSN

In this system, Volvo's earlier system is actually implemented to a certain extent ahead of schedule. Volvo will later make the entire intelligent driving, cockpit and vehicle control into a large VCU, and then connect it to the VIU.

Figure 6. VIU architecture

● VIU is similar to a regional gateway, and the nodes below are all mechatronic Rim nodes;

● Mechatronic Rim nodes are beginning to show very specialized intelligent control, and even single-chip, similar to the complete LIN solution with control released by drive manufacturers before;

● VCU is the top-level logic unit that Jidu wants to build. It controls the lower-level network through SOA and provides vehicle-level control behaviors (vehicle power control, air conditioning control, external lighting, and internal lighting).

We can understand why Jidu was able to enter the centralized architecture so quickly. It is indeed based on the fact that the lower-layer network of the entire vehicle has been doing SOA work on a common platform and connected to you via Ethernet. Then you can focus on the two main software units of automatic assisted driving and cockpit.

Figure 7. System architecture

summary:

I personally think that the case of Jidu has made most vehicle manufacturers rush into SOA and centralized architecture in 2023. With the massive introduction of lower-level gateways, this model of making software if you want to make software and sharing the cost of the overall architecture is indeed a trial and error model. Everyone is still competing for the experience of the entire software set, and this time point will officially start at the end of 2022-2023.