10.5.23-Showcase-Tokyo-iSee

YIBIAO ZHAO: Hello, everyone. I'm very glad to be here to present our company. I'm Yibiao. I'm CEO and cofounder at ISEE. At ISEE, we are building autonomous driving solutions for logistics yards. And before we started a company, I was working with my colleagues, my co-founder Chris Baker and Professor Josh Tenenbaum, at MIT. And we are studying collaborative robots. So the central question we're trying to answer is how machines can learn and think like humans, especially interact with humans in a human environment.

So here is a very interesting example. Today, we will talk about machine learning, we talk about learn from big data or robots follow the rules. But the kids, like 18 months old without anyone tell him to anything, there's very interesting behavior emerging fully autonomously. So we can see the 18-month-old is observing an adult trying to accomplish something. But apparently the adult cannot accomplish his task. The 18-month-old, without anyone told him anything, he opens the door for the adult. And look at him, confirming am I helpful? Did I did the right thing? So that is very impressive compared to what the robot can do today.

And you may ask why this is important. So today, most of the robots applied in the real world pretty much just follow predefined rules. And they lack the basic common sense to understand the human behavior and how to interact with humans appropriately and safely. And that is our vision. At ISEE, we're building this advanced humanistic AI. And we apply it to modernize the global supply chain.

And when we look at all the applications through the autonomous driving spectrum, we pick up this yard use case. Yards are the logistics hubs, warehouses, ports, or airports. There's a lot of application that those trucks only operate in a private road. And therefore, there's no public road regulation. There's no insurance challenges. And the problem is fully under control in a private space. It's also fenced area. There's no random people. There's no kids. There's no wild animals. It's just much more well-defined problem.

But that doesn't mean that it's easy. There's a great shortage of qualified drivers because driving a truck in a very congested space is hard. But not only that, you can see, this is a typical intersection in a yard. There's no lane line. There's no traffic lights to regulate the traffic. It's completely unstructured environment.

This is a video I took in my first time visiting a customer site a few years ago. And I was fascinated by this video. So this is just the daily operation. And multiple trucks come to the intersection. The question is, who should go first, right? It's like this 18-month-old interaction, that we take that almost as granted. Right, it's so easy for human, but it's not trivial for robot to figure out this complex interaction.

And then, starting from there, we built a system that learned to negotiate without rules. Like we mentioned, in the yard there are no rules. And basically, those trucks need to figure out how to interact with each other in real time. So on the left, So you can see that's a simulation engine we're running. And those orange things are static trailers from the sensor data. And we simulate all those trucks those are truck only. Those are tractor trailer. A large scale simulation, like hundreds or thousands of trucks across multiple customer sites. And we simulate how they interact with each other by applying reinforcement learning, game theory, cognitive modeling.

And here's a very interesting interaction. Here, this truck is a little truck come to this. Oh, there's confliction. And this guy learned to make some space for the big guy so they can both go through seamlessly. That's just like this 18-month-old, right? So there's the simple intuition, like how you interact and unblock the traffic in real time. And when we apply it to the real world, this is a real video of multiple trucks coming to the intersection. You can see one aisle is narrower than the other side. So both trucks decided to say, OK, this truck should make some space for another one. And I want to say that both of those information, those rules are not coded to the system. We didn't say the small ones have to yield to the big one or the one in the wider space have to yield to the narrow aisle. All those are emerging from the learning on the fly. And that is really fascinating.

And that's the core to solve a general autonomous driving in a complex human environment. But when we put everything together, this is how it looks like. Here is our autonomous driving truck that's operating in one of our customer sites. So what is unique about this video is that we let the driver, safety driver, step out of the cab. So it's truly driverless. And we're not just driving. We also handle the other tasks, including find where's the trailer, couple with the trailer, and then raise the boom, connect the air lines between tractor trailer, so you can release the trailer and release the trailer brake and pull the trailer out of the spot, so driving fully autonomously to the target location.

I want to point it out, this is not a demo, not a pilot. We have done this, multiple trucks in multiple customer sites, and use them in a daily eight hours, full-shift operation. And we have moved more than 10,000 loads for our customer in daily operation.

I just want to point out that customers compare a lot of different metrics, including safety, including efficiency, and our system perform those tasks comparable with human drivers, and sometimes even more efficient. For example, backing a trailer into a dock door is not trivial because trailers comes with different length, different wheelbase, different shape, different weight. And it's inherently a challenging, unstable system to control. But we consistently be able to back into a spot in one shot and often better than most of the human drivers can do. And that saves time.

So today we have partnered with some companies across multiple spectrums, including warehouse, storage, distribution, logistics, freight, port, and transportation. We especially have a Japanese car OEM customer, that we helped them to move 50 trucks within their big manufacturing plant in the US. So we're looking for more strategic partners, with working with Asian companies. So if you're interested, feel free to reach out to me in the booth. I'm happy to talk more.

That's it. Thank you very much.

[APPLAUSE]