使生活在道路上更容易为司机与 AI 灌输的自主权
这篇文章来自 nvidia.com。原始 url 是: https://blogs.nvidia.com/blog/2017/12/28/trucks-ai-autonomous-trucks/
以下内容由机器翻译生成。如果您觉得可读性不好, 请阅读原文或 点击这里.
Episode 2 of “I am AI” docuseries explores safer conditions for truck drivers.
Truck drivers haul most of the freight that traverses the U.S. With long hours, mounting demand to fulfill online orders, and the perils of navigating everything from snowstorms and to rush hour traffic, it can be a stressful job.
At its technical center in the northwest corner of Washington state, the 25,000-person, $17 billion-a-year company is making advances in automating semi-trucks — with the potential of providing relief for overworked drivers.
PACCAR has recently opened an innovation center in Silicon Valley that will coordinate next-generation product development and identify emerging technologies that will benefit future vehicle performance.
“Trucks really are the lifeblood of the U.S. economy,” said Carl Hergart, director of advanced technology at PACCAR. “What we want to do is make the truck driver as safe, comfortable and productive as possible.”
Helping Trucks “See”
As part of their research into highly automated systems, PACCAR engineers are placing cameras and sensors all over truck exteriors — on grilles, bumpers, mirrors and the like — and then using the NVIDIA DRIVE PX 2 AI supercomputer to crunch the voluminous data collected. The fused image data that’s processed by the DRIVE PX is then run through a neural network that helps the truck understand what it’s “seeing.”
The driver engages the autonomous driving mode with the simple flick of a switch in the cab. Under the hood, a whole set of technologies start to feed the truck a constantly evolving map of its surroundings.
“The truck is building a picture of the environment from the sensors,” said Christopher Balton, senior powertrain controls engineer at PACCAR. “Based on its position in that environment, it makes decisions on where it’s going to drive.”
Benefits of Transfer Learning
What’s really helped PACCAR’s efforts is a concept known as transfer learning. This enables the company to seamlessly tap into the training data NVIDIA has assembled using cars. And it’s an important consideration given the impracticality of trying to map the nation’s roads using trucks.
“We were able to take neural networks trained by NVIDIA in a completely different environment and apply them directly into our vehicles,” Balton said. “Being able to transfer that learning into our environment has a lot of power.”
Not having to create that data itself has also freed up PACCAR to power ahead with more advanced development. For example, it’s currently working to making the Level 4 system more robust to poorly marked roads and inclement weather.
Drivers Not Going Anywhere
As is the case with most AI development, the promise of autonomous trucks has stirred fears that the technology could one day replace drivers altogether.
“What we’re looking to do is absolutely not replace the driver, but to make a tough job easier,” Hergart said. “If we just look at the aviation industry, we’ve had autopilots for a very long time, and we’ve still got two pilots in the cockpit. I think it’s going to be a similar story for trucking.”
观看更多的情节 "我是 AI" docuseries 对我们的 AI 创新者页面.