On-Board Computing Cannot be Replaced by 5G, Says NXP

Paul Myles

TU-Editor

---

29th January 2021

In vehicle computing power will grow exponentially well in advance of the dreams of sending mission-critical autonomous function data to the cloud.

That’s the view of Arnaud Van Den Bossche, director, global product marketing for eCockpit and ADAS, at NXP Automotive Processing. While many eye the possibilities of 5G magically advancing upper level autonomous technology, Van Den Bossche says this is a long way off. He believes the vehicle’s independence in mission critical decision making is essential to ensure adequate safety.

Latency

While 5G promises greater bandwidth and improved latency than the existing 4G networks, he sees the cloud’s role as one in ‘training’ autonomous vehicles by collecting data for future use rather than in making real-time decisions for the vehicle. Van Den Bossche said: “What we do see is that there is a lot of computing power required inside the car. Of course, in the training of autonomous cars a lot of data will have to be pushed from the car to the cloud in order to have the data from all the different cars that will have to be trained.

“Then, when all the training is done, it will be pushed back to the car itself. However, right now to do the processing power when you are driving in a city or on the open road, to then try to push the data from the sensors to the cloud, processed in the cloud and then pushed back to the car – no, this cannot happen.”

Van Den Bossche outlined the issues that 5G will not be able to solve over the existing problems any long-time motorist can recount despite the explosion of infrastructure that 5G will demand to provide adequate cover for even basic connectivity. He explained: “For certain reasons, like if you are in a rural area with not a lot of coverage you will start to have a bottleneck for the data. Or you have the exact opposite where you are in a very dense city with a lot of tall buildings and other cars around, then again you are creating another bottleneck.”

Data compression

While these bottlenecks may not be so severe as the dark-spots for coverage experienced currently in remote rural locations or in the ‘canyons’ between high-rise city buildings, the sheer speed required for decision-making will not be met in the cloud. He said: “Even with 5G it will be difficult to have all the data in real time because, here it is important not to forget, everything with driving a car has to be done on a real-time basis. There is also the safety question that are we sure that everything that is going to the cloud and then coming back is accurate. For example, we do not use any compression of the data from the sensors right now in order to have very good accuracy and all the images. If you are transferring to the cloud, then you have a lot of data and limited bandwidth so some compression has to take place.

“So we think that you will always have to have a lot of computing capabilities inside the car for safety purposes to ensure all the accuracy of the elements needed for recognition in real time. For example, if you are looking at a camera that is doing 8 megapixels, 60 frames per second that you have on a front facing camera, then you have you have another four cameras around the car including the rear facing camera, plus you have the radar the data from which you have to stream. So, if you’re looking at all this data from all these cars, it’s at least a megabyte of data that you have to stream outside of the car in real time.

“Imagine having to do this in a dense city like Hong Kong with very tall buildings, you will have problems with transmission including shadowing problems depending on where all the antennae are placed. Of course, milliseconds count when you are driving at 50kmh (31mph) because the range we take to make a decision is within the 10-15 millisecond so you cannot have any delay whatsoever.”

Van Den Bossche believes there are several near term advantages of large capacity computing capabilities on board current connected vehicles. He suggests: “For me there are three phases and the first phase, from my point of view, is about to be solved it’s about having enough performance to do two things. First to take on the perception of the model of the car and the environment it is in. Second is to make a decision, such as where do you need to go? Here the important thing is the car understanding what will happen at T+1 after making the decision.

“It’s difficult to understand how much processing power will be needed in this area because it requires new algorithms. Right now, perception is fairly well understood using LiDAR, radar and camera sensors, depending on whether the car brand is more dependent on camera or radar sensors but there will be a mix and match of the two to make sure you have a good understanding of what is around you. Whereas the decision making has to handle the prediction about all the objects that are around the car and this is a bit more on an unknown quantity.”

Level 2+

“Now the second phase is about safety and here we need highly integrated perception to be able to achieve highly integrated decision making. Without safe perception where you cannot trust what is happening to make the right decisions. That’s where we are, I think, are trying to solve right now and the reason we are stopping at Level 2+. Because here, if there is a problem, the handover goes automatically to the car driver who must know his/her own environment.”

Van Den Bossche warns that the level of liability facing automakers and suppliers is one of the main challenges in advancing past Level 2+. He explained: “In this way the automaker and supplier escapes too much liability because the drivers owns the environment and can override any kind of decision. When you get to Level 3, there is this lag time of five or ten seconds in which time a lot of decisions have to be taken. In that time, the driver needs to be told something is wrong but the car still needs to be driven. This we do not know entirely how to handle the handover. No regulations have yet been set out for this level and it’s a super complex situation.”

— Paul Myles is a seasoned automotive journalist based in London. Follow him on Twitter @Paulmyles_