10.10.23-Showcase-Seoul-iSee

YIBIAO ZHAO: So unfortunately, I tested COVID this morning. So I cannot be there in person, but I'm very glad to present the company in Korea to you. So we're iSee. We're working on autonomous driving solution for the logistic yards. I'm Yibiao. I'm the CEO/cofounder at iSee. Before I started the company, I was at MIT, working with Professor Josh Tenenbaum and Chris Baker, who is our chief scientist at MIT. We studied how to build machines that can learn and think like people.

So here's a little teaser. Today when we talk about AI or robotics, we often talk about learn from big data or the programs that follow specific rules. But let's look at what human can do. In this video, there's 18-months-old sitting in a corner. Without anyone telling him anything, he's just observing what the adult's trying to do.

So this guy-- apparently trying to accomplish something, but he cannot. So look at this 18-months-old. He slowly walk over, open the door for the adults, and confirming, am I helpful? So that's very amazing because as humans, we often take for granted all those [? innate ?] capabilities that we have-- but how much a young kid can understand the intention about others and seamlessly interact with others in the social environment.

And that is what we needed for robots, or autonomous driving vehicles, in general. So because of that, we decided to start a company to build this humanistic AI that applies to global supply chain.

And while we're doing that, we're also looking at where we should apply the technology, and that led to us to the yard, the logistic yards. So unlike the other use cases, like robotaxi or long-haul trucking, logistic yard is a private lot. It doesn't have the public road regulation. It's also a fenced area. There's no people, like random people-- no kids, wild animals. It's a very well-defined environment.

But it doesn't mean that it's simple. As you can see in this video-- this is a video I took my first time visiting our customer site. And when multiple trucks come to intersection, it's not trivial to tell who should go first, just like the 18-months-old standing in the corner. So it's not easy without clear rules to say who has the priorities-- that people have to figure out their plan and priorities on the fly.

And then plus, you can see there's no lane line, there's no traffic lights in [? the ?] environment. It's completely unstructured environment. So that's why it's challenging to build an autonomous system in such an environment. In the past years, we built a system that learned to negotiate without rules.

As you can see on the left side, this is a typical yard-- a customer yard looks like. The orange ones are the static trailers, [? the ?] park the trailer. And we simulate a large scale of tractor trailer moving back and forth in the yard for millions of iterations in the cloud. And over time, the agents start to learn the autonomous behavior very intelligently, just like 18-months-old. Just pay attention to this area.

So this little guy, he made another big tractor trailer, and he made a move. He decided to make some space and then [INAUDIBLE] its own way. It's very [? intelligence ?] behavior. I want to mention that there's no rule that coded in the system. It's intuition learned by the system, how to solve those puzzles on the fly.

On the right side, that's a similar but slightly different interaction. Two trucks come to an intersection. It's not very apparent who should go first, but because one side of the aisle is narrower and the other side is wider, both of them decided to say, oh, the truck in the wider space should make the move.

And I want to mention, there's no rules included in the system, and the system over time learn those common sense that'll be able to navigate seamlessly, like human in a complex environment. When we put everything together, this is how it looks like. This is one of our [? truck ?] operating in our customer site.

So what is unique about this video is, we ask the driver to step out of the cab. So it's truly driverless, without human involved in any step of the automation. So the truck is not just driving, it also accomplished additional steps to pick up trailer, drop off trailer. Especially, you can see [? that ?] involved a [? truck ?] test, raise the trailer boom, connect the [? airlines ?] so it can release the trailer brake and then pull the trailer out of spot and navigate to the target destination.

So our customer has evaluated us both in terms of safety, efficiency. And our system performed comparable with the human performance. And as you can see, there's also other human drivers in this space. So our system can navigate in a mixed traffic environment very well.

Here, once we get to the final destination, our vehicle is trying to align itself and backing to the dock door in one shot. As you may know, parking a trailer backwards is very challenging, technical task, but our system did it very well. And 90% of the time, it's a one shot, and often, better than what a human driver can do.

After we're parked at a target spot, we drop the trailer, disconnect the [INAUDIBLE], and drive away. That finish the [? move. ?] Yep, that's the brief intro of our autonomous driving technology for the logistic yards. Unfortunately, I cannot present in the booth, but if you're interested in our technology, feel free to email me at yz@isee.ai, and I'm happy to continue the conversation and tell you more about iSee. Thank you.