NVIDIA's smart city vision: AI police cars, AI trucks, AI robots and more
Building the stepping stones of an AI city through deep learning, AI robots, smart surveillance and more - NVIDIA shows the future of cities with AI enablement at GTC 2017.
Just a few years ago, deep learning and artificial intelligence (AI) were still the stuff of dreams. Today, the increased processing capability at edge devices and the cloud, combined with high speed connectivity are enabling distributed computing, machine learning and deep learning that all come together to deliver some aspects of AI. AT NVIDIA's GTC 2017, this is exactly what the company wants you to take away: the AI city is coming and is enabled by NVIDIA's building blocks and from its partners. We roundup some of the show floor highlights that best exhibit NVIDIA's involvement in the AI city future.
AI Trucks
What’s a truck doing at a graphics technology conference? That’s because NVIDIA is developing autonomous trucks with trucking giant Paccar to make them safer, smarter and efficient with the help NVIDIA’s edge processing that’s powered by deep learning.
Paccar manufactures other lines of trucks such as Peterbilt, DAF, Kenworth and others will all one day benefit from autonomous driving up to Level 4 capabilities, which means the system would be able to handle routine driving operations without intervention of a driver.
The brain of this truck is powered by NVIDIA Drive PX 2, a compact in-vehicle supercomputer built for autonomous driving. The 8 teraflops of compute power it packs can deliver up to 24 trillion deep learning operations a second while processing inputs from up to 12 video cameras and a range of lidar, radar and ultrasonic sensors. What does this all translate to in real life usage? The Drive PX 2 can enable lane keeping functionality, object detection and free space calculation to manage adaptive cruise control or slow down to avoid collision.
The brain of an AI truck:- NVIDIA Drive PX 2.
AI Police Cars
The future of policing a city likely won’t be in the form of Judge Dredd. Instead, AI police cars like this are more likely to patrol our streets. Contrary to what you might be thinking, AI police cars aren’t replacing policemen, but rather, a more intelligent vehicle can enable police officers to do their job more efficiently. With greater insights gleamed by surveillance data gathered by the car’s sensors to detect anomalies or any other concerns pertaining to the open cases that the police are currently investigating, the AI police will be a formidable tool for combating crime, maintaining peace and keeping the public safe.
The wonders of deep learning making sense of real-time surveillance data.
Inside, a quick peek at the control console that could possibly scan the vicinity to track activity and ID people through facial recognition. Again, this is also made possible through deep learning.
Mounted atop the AI police car, a drone that can be dispatched in a moment's notice for remote surveillance.
AI Robots
Robots are almost everywhere these days from manufacturing in factories, to robot vacuum cleaners in the home, but all of these are highly purpose specific with fixed scope and follow programmed routines. AI robots are the next stage that gather and process real world inputs and make decisions in real time facilitated by algorithms trained by deep learning.
Meet the Knightscope autonomous security robot that’s a perfect partner to AI police cars. It’s designed to monitor crime in a set locality by running real-time video analytics that’s powered by NVIDIA Jetson within the robot. It can detect and report people (especially those on the criminal list, on parole, etc.) during restricted times or in restricted areas, perform forensic analysis on license plates and more. The Knightscope robot is equipped with a variety of sensors, range finder and a camera to aid in its data gathering surveillance.
A preview of the robot’s control panel to get a sense of the data that it’s juggling with at any one point of time.
Elsewhere, AI robots like the Fellow Robot (below) is designed for inventory management and customer servicing for in-store product inventory management with the assistance of deep learning at the edge device (the robot).
Looks like this one is working at an NVIDIA store. Now iIf only we could convince it for a 90% discount off the Shield TV...
The hardware behind AI and Deep Learning
With all the practical use case scenarios shared on the previous page, we’re sure you might be curious on what sort of machines and hardware are behind the scenes to pick up all the deep learning efforts and the training required to establish deep neural network models; so here they are in no particular order.
For high performance computing and training of deep neural networks, you will surely benefit from massive firepower such as this Hyper Scale PCIe Node 2U rack from Inspur that has the ability cram 16 NVIDIA Tesla GPUs in one box.
Here’s another look at the same rack; are those Teslas calling out to you yet?
Of course, if you’re looking for the very latest, this is the NVIDIA Tesla P100, the world’s first high performance computing unit for cloud servers as claimed by NVIDIA. It’s indeed several times more capable than a conventional Tesla GPU or even a bunch of them combined.
This rack shows that it can take in up to four Tesla P100 computing units as seen by its NVLink connectors on the motherboard.
At the very top of the chain, you’ll want NVIDIA’s DGX-1, a supercomputer designed for deep learning.
It can be configured with 8 x Tesla P100 compute units! Oh and it costs US$129,000.
On the opposite end of the scale, the NVIDIA Jestson is an edge computing device the processes input data and makes autonomous decisions based on the deep neural network model that was trained prior through deep learning in the cloud using powerful machines.
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.