For the AV sector to continue to make progress the one thing you need is time. And that is actually happening now. The other part that's happening is that the performance computer platforms, from the likes of Nvidia and Intel allows all these deep neural networks (DNN) to come in. It's amazing the progress that's being made at the moment, you can feel it, you can actually feel it moving now.”
“China has the opportunity to build new infrastructure, real physical infrastructure that is almost impossible to build in Europe and the US. We can't even make a change to an airport, but in China they can completely replan cities thinking about automated driving. In the larger perspective, they are really thinking about smart city robotics, delivery vehicles, street sweepers, and now they have these autonomous vending machine vehicles. It's a little bit like when the iPhone was released, nobody actually realized what you could do with apps till the iPhone appeared in 2007.
I think the idea of a driving vending machine is absolutely fabulous. Let’s say you have used your last print cartridge: on the app you ask for a replacement print cartridge and a small vehicle drives one over to you,” he enthuses. “This is a great example of limiting the environment in the operational design domain (ODD), a much more controlled environment where innovations such as this are much easier to implement.”
Lambermont is of the opinion that some of the major automotive OEMs have been all-but forced into reacting to pressure to innovate from the arch innovators themselves, Tesla.
“Actually, I think all the OEMs feel the pressure of Tesla, and shadow mode learning. They are all working on that. And that relates to that earlier question. If you're able to have a central computer platform and you can put all your neural networks in that central computer, you could actually learn new features and validate software in the field instead of having the expensive verification and validation mileage accumulation exercises.
But it isn’t the only enabler, as Lambermont agrees.“We are concerned about the social aspects too. These applications are at their most useful when they tackle tedious tasks, like driving a shuttle at an airport in the middle of the night when there's only a few passengers on board. It doesn’t need a human being to do that. I think everybody understands that we should automate the task of moving elderly people to and from hospital or to and from their retirement/care homes.”
If this year has taught us anything it’s that working together, collaboratively, can bring about variously untold and previously unheralded benefits. Lambermont is quick to point out that collaboration is something very close to his heart.
“This is my dream!” he jumps in, enthusiastically. “I'm actually working on this as a startup called Resembler (www.resembler.ai). When we learn to drive and we pass the test and get our driving license, you can't actually drive. All you can do is follow the traffic rules. We learn how to drive every time we get behind the wheel AFTER we get our licence, for instance when we’re suddenly in a dangerous situation. We remember it, and then in the future we avoid that situation or we at least know how to circumvent it. Humans are extremely good at this. When you overtake a cyclist the distance that you leave them depends on if it is an adult cyclist, if it is a child, if it is windy, if it is icy and that decision is made in a tiny fraction of a second. If autonomous vehicles can learn from these situations then we can accelerate the process of them becoming safer.”
Lambermont concludes with a notion that is undoubtedly shared by the entire autonomous vehicle, and human-driven vehicle sector. “I’ve just described my dream for automated driving, because now you can really use real-world, technical knowledge about what is dangerous in order to make the world safe.”
Comments