This GM Technology Could Create the Future of Self-Driving Cars

With its self-driving vehicles on the road now for several years, Cruise from General Motors (GM 1.13%) has been collecting valuable real-time data from actual driving scenarios that could not be created in a simulation. In this Backstage Pass clip from "The AI/ML Show" recorded on Jan. 12, Motley Fool contributor Jamie Louko talks about how this approach may set Cruise up to succeed in its plans to scale into major markets around the world. 

Jamie Louko: This is just a really one picture that I really like here. It has some of these pictures and RADARs right down here showing what it looks like in front and to the side to the side. But up here, it shows what the car is picking up. Then this green line right here is the predicted direction that the car is going given all of these things here. If you see this line, that shows where this car in particular, but where cars are expected to move. Theoretically, if there was a car coming in through this road, it would have lines coming out of the road showing where the car could go. I think that's really interesting. If you look really closely, it labels everything it sees. It has a person here, a person there, car, car, car, and it's really interesting.

Then they are eventually want to get into some simulation where this is another example of what they're using and how they're seeing everything in front of them. Just a couple of things that I really liked about the business. It's been testing on the streets of San Francisco for two years. I've never been to San Francisco, but watching all the videos of Cruise driving around, I'm really glad I don't live there because it looks like it's a terrible place to drive. What I really like about this real-time data is that they're presented with scenarios that I never have even thought of.

One of the really weird scenarios was there was a trash can flying in the road, three feet off the ground just flying across the road. You wouldn't get that no matter how many times you tried in a simulation. With these real-time scenarios, they are collecting over 100 interactions per scene specifically in San Francisco. They compared it to other suburban areas or highways where they received significantly less. There were talking 30-20 interactions in daily collection points per scene.

They are focused on ride-hailing and delivery experiences for the short term. Over the next 10 years, they have a different vision, but right now that's what they are focused on. When it comes to mapping, that is why they are using it, because it doesn't really matter. As long as they're focused only in San Francisco and only in Phoenix, which is where they're mostly doing their delivery for food delivery, they really don't need to go out of that area. They can set a range that they want and they don't have to go out of that area. Mapping, in this case, really makes sense, at least for the short term.

In terms of other ride-hailing competitors and food delivers, their economics are so much better, they don't have to pay any drivers. With liabilities and things like that, they collect much more profit per transaction than say, an Uber would or something like that.

I want to stop right here. I will go over these in some other scenarios, but this shows what I mean by performance at all costs, they're really wanting this high-cost, they're willing to take it on now as long as they're getting up here. Right here is where their performance is, it's clearly much higher than some of these other competitors. Then they're eventually, once they get this to scale, then they're going to look at decreasing their sensor count and moving more in the more affordable cost. But right now, they're focused completely on performance, and they'll focus about the cost later.

They have been driving without backup drivers since 2020, which I think is really cool. I like their unique approach to their market focusing on that performance, which really gives them the potential staying power in the proof of execution that some players, maybe like Tesla, I'm not going to name names, but that might not have fully yet. They have a bunch of key partners and investors. Obviously, GM's a pretty big one, but also another one that stood out to me was Walmart. They're partnering with Walmart in Phoenix to work on their delivery. Once they get approved in Phoenix, they'll become a full partner with Walmart and they will exclusively deliver groceries, which I think is really cool.

They also have something called occlusion training, which was like 11 o'clock at night. I was watching this video, and they mentioned occlusion training, which is literally assuming the worst and assuming things aren't there until they can prove that they're not there. There was one really cool example where a car was coming up to a T, and you could see on the camera, this right here, they were coming up to this T right here. Until they saw around this corner, they assumed that there were cars getting ready to drive this way until they pulled up here. I think that's really interesting and that mitigates some sort of risk. I'm not sure if other competitors do this. I would assume they do, but I thought that was something really cool that they're specifically focusing on.

But in terms of timeline, the ride-hailing is only in San Francisco right now. It's expected to launch sometime in the first half of 2022, their deliveries is in the second half of 2022. Right now, they only have their first-gen vehicles, but the launch is really expected to be on Origin, which is basically a little bus. I see it and I think of like a loaf of bread. It doesn't have any driver, and instead, it's just a box full of seats where people can be delivered to different places, so it's a ride-hailing service. They are really expecting that once they've launch this little breadbox of a car in 2023, 2024, that's really going to be the launch point of their growth and scale-up. Until then they're only thinking like 10s to 100s of cars on the road. But after that, they are really expecting it to accelerate.

In five to six years, they said they want to scale into some major markets around the world, as well as being mostly in the U.S. and almost all parts of the U.S. By 2030, million car fleet, 8-10 years. "They want to be the world's largest supercomputer and mobile sensor platform," which extends well beyond autonomous vehicles. They want to expand our software, deep learning capabilities into commercial vehicles with their GM partnership. This is where GM really comes in. Part of the reason Origen is going to be launched point it's because it is 60 percent more cost-effective than the first-gen.