Computer Vision News - March 2018

21 Xiaodi Hou - TuSimple The purpose of making revenue is not to make us rich: it is important for us because we want our business to be sustainable. We want to make money so that we can pay our employees. By paying our employees, we can make greater technology faster and also recruit greater talents: money is a very important link in this model. Which technological developments have succeeded and which still need improvements? That’s a very interesting question. I think there are people who, when they talk about autonomous driving, always say that perception is really hard. You can’t have perfect perception. By perception, I mean to see other vehicles and measure the distance between you and other cars. Is this kind of object detection? Yes and no actually. Object detection is one of the specific technologies, but what I’m saying is not a technology. It’s a goal. The goal is to perceive the world. People believe that the perception part is the hardest challenge. That’s the reason of talking whether using LiDAR or cameras. By the way, we use primarily cameras. However, the bigger problem to solve is not the perception, but the prediction problem for understanding the human world. That is a very challenging problem, and it’s going to be a challenging problem within the next few years. It will take a long time to fully solve that problem, but I am confident that we are going to solve it. The reason why this problem is still not in the horizon of many people’s perspectives is because if you’re talking about autonomous driving, the first challenge that you have to solve is perception. Before you solve perception, you will not be able to go into the other problems… for example, predicting other people’s behavior or understanding the human societal norms. That is why it’s a much bigger challenge, and the technology required to solve this problem is not there yet. There are some people talking about artificial general intelligence, AGI. You probably have heard of that. Just a small fraction of people are talking about that. It’s really not there yet. The core technology that would finally solve the AGI problem would be used to solve the autonomous driving problem to make sure that the car or truck will co-exist with humans within the human world. Society requires to obey the societal norms: for example, as a vehicle you have to be polite. Some vehicles are doing something that is predictably dangerous, and we want to get rid of that. Guest Computer Vision News