Le Wagon ApéroTalk: Who will win the self-driving race? with Charly Walther

Le Wagon ApéroTalks are back for season 4! For the first talk of this season, we had the pleasure to welcome Charly, former Product Manager at Uber, now working for Gengo (we had the chance to welcome Matthew, Gengo’s CEO, a while back).

As most buzzwords and hyped technologies (think Blockchain, Fintech…), you hear all sorts of things about self-driving, but very few people understand the ins and outs of the technology and market. Welcoming Charly, we sought to give an overview of how self-driving works, what are the main roadblocks for it to become mainstream, and maybe most importantly: who will win the self-driving race?

How do cars become robots?

Before being able to answer the crucial question about who will win the race, you first need to understand how the self-driving technology actually works. The key element on self-driving car is the LIDAR (Light Detection And Ranging), that weird 360 camera you see on top of each vehicle. It basically works like a sonar, except it uses light instead of sound waves to “understand” the environment around it. There is only one major company producing LIDARs at the moment, Velodyne, which is interesting when it comes to understanding the self-driving value chain. The second major piece of hardware you need to know about is the computer itself, basically a large brick stuck in the trunk of the car, powered by a GPU. The GPU itself differs from a CPU in a way that it’s very good at simple calculations — a lot of them.

The main role of these two pieces of hardware is to perform three tasks, which are at the heart of any automated systems (including Softbank’s Pepper, and that plane you flew last week): Perception, Prediction and Planning. In other words: What are the objects around, what is going to happen, and what am I going to do about it?

What is Product Management applied to self-driving?

Generally speaking, a product manager’s role is to figure out what product should be built. More specifically, the job involves figuring out issues and usages, and dealing with different teams to address those issues. Now, how does that apply to self-driving?

There are actually various aspects that product managers need to deal with, ranging from human experience to performance analysis:

A few fun facts about these different aspects:

• Some companies have teams of hundreds working on only the labeling. The only way you can indeed train your AI is by consistently telling it which is what (this is a tree, this is a stop sign, …), and the best way to achieve that today is… fully manual.
• The simulation part implies recreating entire cities, having an entire test-drive course in your backyard, and yes, someties using Grand Theft Auto as the simulation playground. One of the key aspects is the ability to mimic actual human behaviors: test dummies will do what you ask them, while people will probably panic and act unexpectedly when a car is rushing towards them at 40mph.
• Believe it or not, Phoenix in Arizona is considered the best city to test self-driving vehicles. Why? Long and straight roads, light traffic, and basically no pedestrians randomly crossing in front of you. As Charly puts it “If it’s going to work somewhere, it’s gotta work in Phoenix!”.

Who will win the self-driving race?

Every other month you hear about another high-profile tech company announcing they’re entering the self-driving market: Google is doing it, Uber is doing it, Tesla is also working on it. But to be able to figure out who will win that race, you need to take one step back from the obvious “hardware” providers, and look at the entire value chain:

And the most interesting question might lie in the “Who will own the network” block. It means that self-driving might actually not be a hardware game, but most likely (and sadly) a business development one. Today, car manufacturers like Toyota or Renault produce the car, and I am the one paying for it. But when self-driving becomes mainstream, and I don’t need to really own a car anymore, one big question comes up: who is going to pay for the car that will take me around? Government is one solid option, and particularly the ones that strongly invest in their infrastructure, like Singapore. Another option is that the industry will mirror what happens in the airline industry: third party will purchase cars, and lease it to operators.

From a purely hardware viewpoint, we also see companies betting on one of these two options: Laser will get cheaper (Uber and Waymo are betting on that), or standard cameras will get better (Tesla’s bet). A noteworthy point is that cameras have an inherent incentive to get better — It’s their main selling point.

A few roadblocks ahead

Once hardware and ownership questions are cleared, the last roadblocks will probably be to gain customers’ trust by clearing the safety and ethics aspects. As someone who was working in this industry, Charly’s view was optimistic (and probably a bit biased). The ethics debate might be the most heated and philosophical one, but it gets clearer once you figure out that even though computers are more reliable than human drivers on average and will only get better over time, they still cannot prevent unpredictable crashes, or react in half a second. Basically, just like a normal human being, when it crashes, it just crashes…

Thanks a lot Charly for giving this talk and sharing your views!

To stay updated about upcoming ApéroTalks, join our Meetup group.