The road ahead: Are we ready for an AI driver? – Part 2

Article By : Junko Yoshida

In the autonomous car, AI will advance machine vision systems, while it will also migrate in sensor fusion electronic control units (ECU).

How ready are we now for the day when we depend on AI to drive autonomous cars? Here's the second part of EE Times chief international correspondent Junko Yoshida's interview with Luca De Ambroggi, principal analyst, automotive semiconductors at IHS Technology.

AI for sensor fusion?

Where will AI technology be applied inside a vehicle–beyond vision?

Cameras are the first obvious place. But I can assure you that there are tier ones and car OEMs in Europe looking at applying AI for radars. AI will play an important role in sensor fusion.

Only seven months ago, Toyota announced a five-year, $1 billion R&D effort on AI headquartered in Silicon Valley. Toyota announced that it will initially have a laboratory adjacent to Stanford University and another near M.I.T. in Cambridge, Mass.

It’s no secret that practically everyone including General Electric, Baidu, Samsung, and all the major automakers has been establishing research outposts in or near Silicon Valley take advantage of its engineering talent. So where are we today on AI technology advancements?

We’re still in the research level at universities and research institutes. My guess is that it will take next five to 10–or more likely 10 years–for the fruits of their research efforts to meet commercial/engineering needs.

How do you certify AI?

Assume that we will reach the point where autonomous cars will fully take advantage of the AI in 10 years. What challenges will be waiting then?

One of the major challenges, I suspect, will be the certification of AI. Just as we certify a human driver [with a driving test], the [automotive] industry needs a set of standards or procedures they can use to certify AI–for safety.

But for that, the automotive industry needs to know what to test in the AI system. I assume that a sheer number of hours driven by a Google car in Calif. and Texas wouldn’t prove the safety of the self-driving car.

Tier ones and OEMs need to develop a set of unified tests–safety parameters–to certify AI. It won’t be an easy task. At this point, no industry alliances or ISO bodies that I know of have approached this subject.

AI is a complicated problem. I know everyone–every tier one, OEM–is working on AI. Many of them today are saying that they would start small [with AI research] now, in hopes of being ready when all the circumstances surrounding AI get in order.

Smoke and mirrors

As the way I see it, there’s a lot of smoke and mirrors when it comes to actually doing AI in automotive chips.

You explained earlier in this conversation how Nvidia’s GPU-based platform fits into an automotive market looking for AI solutions. But you did say that their platform is more for testing AI in vehicles, rather than implementing AI in mass production models. Who else is in the running?

Tesla, for one, was said to be doing some AI when it enabled autonomous driving features—by using Mobileye’s EyeQ3. But I wouldn’t call that deep learning.

My understanding is that Audi's 2017 Q7 model also has the capability to run AI-based algorithms, if Audi decides to do so. But I bet they won't deploy it by 2017. We aren’t sure if that’s using Mobileye’s Q3 or Q4 chip, and how much AI is being used in that model.

The bottom line is that a lot of AI claims are being made today by chip vendors and car OEMs. But for genuine deep learning, you need a big muscle in your processor.

Aside from Nvidia and Mobileye, we suspect that NXP Semiconductors is naturally working on AI.

Xilinx and Altera/Intel are also big on AI. IP vendors like Ceva, Synopsys, Cadence and Mentor Graphics are also going after the AI space.

Take a look at recent merger and acquisition deals in the automotive technology market. There’s a lot going on.

[Editor's note: Just in the last two months alone, Intel acquired two companies related to the field of ADAS, robotics and autonomous machines. In April, Intel acquired Yogitech, an expert in semiconductor functional safety and related standards. In May, Intel announced the acquisition of Itssez Inc., a company armed with Computer Vision (CV) algorithms and implementations for embedded and specialised hardware.]

__Figure 1:__ *AI-based systems for automotive (Units: Millions)
(Source: IHS Technology)*