Smart Cars Need Smart Roads, and PerceptIn Believes It Can Deliver Exactly That
Envisioning a three-stage rollout, the company claims that future autonomous vehicles could offload much of their work to infrastructure.
Jean-Luc Gaudiot, professor of electrical engineering, and Shaoshan Liu, founder and chief executive of autonomous vehicle startup PerceptIn, have put forward the argument that smart self-driving cars aren't enough — unless they're paired with smart roads, too.
"For now, at least, there is only so much sensory power and intelligence that can go into a car," the pair write in an opinion piece for IEEE Spectrum. "To solve this problem, we must turn it around: we must put more of the smarts into the infrastructure — we must make the road smart."
They have a point: over the years we've seen a range of autonomous vehicles, from insect-inspired gas-sensing swarms to farming robots, roofing drones, bird scarers, and miniature food delivery trucks — but fully autonomous self-driving vehicles capable of handling the same road conditions as a human remain elusive.
The solution: better communication between both the vehicles and the surrounding infrastructure and between individual vehicles themselves. "Here’s how it could work," the pair write. "It's Beijing on a Sunday morning, and sandstorms have turned the sun blue and the sky yellow. You’re driving through the city, but neither you nor any other driver on the road has a clear perspective."
Each vehicle, though, can provide a snippet of information, which a centralized system can use to build an overall view with far more detail than any individual vehicle on the road — and combine that view with other data sources, like traffic cameras and weather services.
To prove the concept, PerceptIn has been working on model deployments — including a 15 kilometer stretch of three-lane highway, going both directions, with a base station every 150 meters. Each base station, the company explains, is an Intel PC with an NVIDIA GeForce 1080 Ti graphics card for GPU offload along with LIDAR, visible-light camera, and radar sensors, plus a communications unit that can communicate with suitably-equipped vehicles.
Broad deployment, the pair admit, is challenge — for reasons both technical and political. They envision a three-stage rollout: Augmented autonomous driving using vehicles with substantially similar sensor packages to today's implementations; guided autonomous driving, which offloads computationally expensive tasks to the infrastructure to reduce vehicle cost; and planned autonomous driving, where the vehicles would have "only very basic sensing and computing capabilities" of their own.
Liu and Gaudiot's article is now available to read on IEEE Spectrum.