This article was created in collaboration with the sponsoring company and our sales and marketing team.
Huawei is making automobiles drive smarter and safer thanks to its advanced driving system called the General Obstacle Detection network. The network combines signals from different sensors like cameras, mmWave radars and lidars, which in turn allows vehicles to truly see and understand different objects.
Supported by the General Obstacle Detection network, vehicles can identify not only common objects, like people and vehicles, but also irregular obstacles that they have never seen before like overturned vehicles, crates and fallen rocks or guardrails. This network helps build a security perception system that makes driving. Thanks to this breakthrough innovation, vehicles can sense the obstacles in their surrounding environments better than human drivers.
Q: Tell me a bit about yourself.
Su Peng: My name is Su Peng. I'm 30 years old. I have a Ph.D. in Electronic Information Engineering from the Chinese University of Hong Kong. My main research interests are computer vision and artificial intelligence. Simply put, it's about helping machines see and understand the world.
Q: When you got your PhD, what made you choose Huawei as the company to work for?
Su Peng: Well, first of all, I've always dreamt of creating a robot as intelligent as a human — and intelligent vehicles, which are intelligent mobile terminals, are the perfect opportunity for me, and Huawei has a wealth of expertise in the R&D of intelligent automotive technologies. So, I think it was a sensible decision to join Huawei.
Q: You're working on something called the General Obstacle Detection network. Can you tell me more about that?
Su Peng: Yes, with the General Obstacle Detection network, vehicles can identify different types of objects and make inferences like human drivers.
Q: It can start learning about different items by itself. Is that correct?
Su Peng: Right. Traditional AI detection models are based on 3D bounding boxes, which rely on manual annotation. But these boxes struggle with representations. Some irregular objects, like a big truck in the middle of turning, two-car vehicles, fallen guardrails, and so on, cannot be represented with 3D bounding boxes, which are usually rectangles. Our General Obstacle Detection network is different. It uses voxelization, which means voxel grids, to represent the physical world in 3D. This can accurately represent different object locations. Our General Obstacle Detection network eschews traditional manual annotation and teaches vehicles how to identify vehicles and people. Our hundreds of millions of kilometers of driving data is fed into the General Obstacle Detection network, which performs self-learning. So far, the amount of data trained has increased by two or three orders of magnitude over the old model.
Q: What do you think the biggest advantage autonomous driving will bring to people?
Su Peng: AI drivers are not hindered by emotion and exhaustion. Studies show that vehicles with AI drivers are on average less likely to have collisions than vehicles with only human drivers.
Q: Tell me a bit more about that safety enhancement. Is there a high point in the development you can talk about?
Su Peng: What left the strongest impression on me was the beta test of the General Obstacle Detection network. When I was conducting my first drive test, a dog rushed out suddenly. If a traditional AI perception system had been used it would not have been able to identify that it was a dog. But our General Obstacle Detection network accurately sensed that it was a moving object and triggered active braking.
Actually, I think it's really interesting that it’s just phenomenal to think what it might be like in 10 or 20 years or what my grandchildren will experience. It will be a very different world than it is today — like a very different world. As a researcher in computer vision and AI, I can clearly feel the speed of technological advancement. I believe that a fully connected, intelligent world is the way forward. This will free us from having to do mechanical and repetitive work, allowing us to focus on the arts, science and tech, which will require human creativity, and which I think will reflect the value of humanity.