NVIDIA introduces DRIVE PX 2 platform for autonomous driving
06 January 2016
At CES 2016, NVIDIA introduced NVIDIA DRIVE PX 2—a high-performance computing platform for in-vehicle artificial intelligence applied to the complexities inherent in autonomous driving. DRIVE PX 2 utilizes deep learning on NVIDIA’s most advanced GPUs for 360-degree situational awareness around the car, to determine precisely where the car is and to compute a safe, comfortable trajectory.
DRIVE PX 2—which delivers processing power equivalent to 150 MacBook Pros—uses two next-generation Tegra processors plus two next-generation discrete GPUs, based on the Pascal architecture, to deliver up to 24 trillion deep learning operations per second, which are specialized instructions that accelerate the math used in deep learning network inference. That’s more than 10 times the computational horsepower than the previous-generation DRIVE PX.
| Jen-Hsun Huang, co-founder and CEO, NVIDIA, introducing the DRIVE PX 2 at CES 2016. Click to enlarge. |
For general purpose floating point operations, DRIVE PX 2’s multi-precision GPU architecture is capable of up to 8 trillion operations per second—more than four times more than the previous-generation product. This enables partners to address the full breadth of autonomous driving algorithms, including sensor fusion, localization and path planning. It also provides high-precision compute when needed for layers of deep learning networks.
DRIVE PX 2’s deep learning capabilities enable it quickly to learn how to address the challenges of everyday driving, such as unexpected road debris, erratic drivers and construction zones. Deep learning also addresses numerous problem areas where traditional computer vision techniques are insufficient, such as poor weather conditions like rain, snow and fog, and difficult lighting conditions like sunrise, sunset and extreme darkness.
| The liquid-cooled DRIVE PX 2 unit. 12 CPU cores, Pascal GPU, 8 TFLOPS, 24 DL TOPS, 16 nm FF, 250 W. Click to enlarge. |
Self-driving cars use a broad spectrum of sensors to understand their surroundings. DRIVE PX 2 can process the inputs of 12 video cameras, plus LiDAR, radar and ultrasonic sensors. It fuses them to detect objects accurately, to identify them, to determine where the car is relative to the world around it, and then to calculate its optimal path for safe travel.
This complex work is facilitated by NVIDIA DriveWorks, a suite of software tools, libraries and modules that accelerates development and testing of autonomous vehicles. DriveWorks enables sensor calibration, acquisition of surround data, synchronization, recording and then processing streams of sensor data through a complex pipeline of algorithms running on all of the DRIVE PX 2’s specialized and general-purpose processors. Software modules are included for every aspect of the autonomous driving pipeline, from object detection, classification and segmentation to map localization and path planning.
NVIDIA delivers an end-to-end solution—consisting of NVIDIA DIGITS and DRIVE PX 2—for both training a deep neural network, as well as deploying the output of that network in a car. DIGITS is a tool for developing, training and visualizing deep neural networks that can run on any NVIDIA GPU-based system—from PCs and supercomputers to Amazon Web Services and the recently announced Facebook Big Sur Open Rack-compatible hardware. The trained neural net model runs on NVIDIA DRIVE PX 2 within the car.
Since NVIDIA delivered the first-generation DRIVE PX last summer, more than 50 automakers, Tier 1 suppliers, developers and research institutions have adopted NVIDIA’s AI platform for autonomous driving development.
Using NVIDIA's DIGITS deep learning platform, in less than four hours we achieved over 96 percent accuracy using Ruhr University Bochum’s traffic sign database. While others invested years of development to achieve similar levels of perception with classical computer vision algorithms, we have been able to do it at the speed of light.
—Matthias Rudolph, director of Architecture Driver Assistance Systems at Audi
The DRIVE PX 2 development engine will be generally available in the fourth quarter of 2016. Availability to early access development partners will be in the second quarter. During the press conference, Huang announced that Volvo will be the first automaker to deploy DRIVE PX 2.
In a public trial of autonomous driving, the Swedish automaker next year will lease to customers 100 XC90 luxury SUVs outfitted with DRIVE PX 2 technology. The technology will help the vehicles drive autonomously around Volvo’s hometown of Gothenburg, and semi-autonomously elsewhere.
AdasWorks worked with Volvo to help create a system that processes data from multiple sensors in real time to provide 360-degree detection of lanes, vehicles, pedestrians, signs and more, enabling a variety of autopilot functions.
Tesla also uses Nvidia but a very old Tegra 3 processor that was state of the art in 2012 when Model S came to life. Hopefully we will see Nvidia's new baby in the next autopilot hardware upgrade for Model S and X.
As I see it the signs are getting more obvious every year that autonomous cars are coming. You have now got autonomous vacuum cleaners and lawnmowers and Space X just succeeded making an autonomous space rocket. Wars are increasingly fought with largely autonomous drones. I read 98% of Tesla's customers for Q4,2015 ordered the autopilot option which is a 2500 USD option. Autopilot is an extremely wanted feature. I expect a premium autopilot option probably costing 10,000 USD may come in the future that can pay for a much more powerful autopilot solution with the hardware needed for fully autonomous driving.
I also expect rapid development of autopilot capabilities in the coming years and full autonomous driving by 2020 for Tesla's vehicles but probably as late as 2025 for everybody else. After that it will be nearly impossible to sell a car or order a taxi that isn´t fully self-driving. Tesla's current autopilot is hardware version 1, and the software is still beta version 1. It is like the iPhone 1 with beta iOS1. It totally sucks by today's standard yet that first iPhone was released just 7 years ago. I expect Tesla's autopilot to progress equally fast because the autopilot sensors and computers will be updated on an annual basis and the software can be updated on a continuous basis using OTA updates which Tesla is still the only car maker in the world that has for its autopilot system.
I do not see battery tech develop as quickly as autopilot systems and I doubt we will ever go much below 100 USD per kwh. Once Tesla's 50Gwh factory is fully operational in 2020 Tesla will be at 130 to 150 USD per kwh at the pack level and it will get very hard to go further down in price going forward. The self-owned BEV does therefore not make sense for mass market adoption as it will need a 100 kwh battery pack costing at least 10,000 USD to have the all weather range that everyone can accept as a no compromise alternative to a cheap gasser costing 15,000 USD. So such a 100kwh BEV will cost at least 25,000 USD and still not be better than the 15000 USD gasser. In other words, owner BEVs is likely never to become mass market cars.
The only economically viable solution is to share the high capital cost of the BEV among many and that implies autonomous BEV taxis that can also save cost by being right sized. If you need transportation for one that is exactly what you get when you order such a ride using your Smartphone. Expect 15 to 20 cents per mile or about half the total cost of owning and driving a cheap gasser.
My excitement for autonomous cars is high because it will end oil and all the pollution that follows with it and it will also save most of the 1.2 million people that are killed every year in traffic accidents and millions more that are severely injured. Plus people will save an hour or more per day that can be used for better things than driving a car.
I really cannot see any other technology that is equally important and that is so close to happen.
Posted by: Account Deleted | 06 January 2016 at 03:47 AM
Henrick,
I hope you are wrong about the 10k price at the same time suprised that the current offering is so cheap.WE are used to seeing capability improvements in the order of magnitude for the same cost or less as a wage percentage."autopilot option probably costing 10,000 USD"
Over the few years you are suggesting, wages may not increase at all except for the top earners and chattering classes.
They will probably have several vehicles garaged at each of the properties but as they are rarely driven they don't represent so highly in accident statistics.
You then say.
BEV does therefore not make sense for mass market adoption as it will need a 100 kwh battery pack costing at least 10,000 USD" Again a cost that makes a BEV unviable is comparable to the cost of a state of the art or current and up to date autonomous option?
then
" a cheap gasser costing 15,000 USD"
Regulation will insist the gasser at 10k is as clean and as safe as the reg's require that won't be optional - even for VW.
Ditto autonomous drive systems - even Tesla.
At the start of the last century Henry Ford understood that if his workers could no afford to own the product they create, that he would not have a market.
I can agree with much of your 'facts but together they don't add up.
" pollution that follows with it and it will also save most of the 1.2 million people that are killed every year in traffic accidents and millions more that are severely injured. Plus people will save an hour or more per day that can be used for better things than driving a car."
Remember the roads are owned and shared by all of us so the 'well off' are not exempt from the statistics. The rest will still need to get to work and will be commuting longer distances so the elderly, disability, other impaired will either represent an increasing hazard or be forced to walk?
I make the same old argument. Is the fence around the Asylum / Prison / gated community there to keep the undesirables out or in? How are everyone’s interests best served?
Finally With regards the Autonomous future I do think it will be possible for A.I. to compete with the highest skilled operators on our road. I won't be an early adopter though
I will wait for the capability to "safely land a passenger plane on the Hudson" version.
Posted by: Arnold | 06 January 2016 at 12:36 PM
The first fully autonomous options for cars will cost at least 10k USD because they need to be redundant meaning two or three of everything related to that driving system. The 2.5k USD option for Tesla's current autopilot system is less because it is not redundant and it only has a small fraction of the high quality sensors and supercomputers needed for full autonomy.
Don't worry about the minimum 10k USD markup for BEV drive and another 10k USD markup for fully autonomous drive. 90% of the population will skip ownership of cars and just use autonomous taxis, buses and trucks instead for all of their transportation needs. They will be cheaper than private ownership because they are shared among many, because they have low fuel costs using electricity instead of gasoline, because they are right sized for the transportation job in mind, because they have low insurance cost as they nearly never are involved in accidents and because they as BEVs are 5 times more durable than gassers. 90% of the population will use them and never own a car or acquire a driver license because they are by far the easiest, safest and least costly way to go anywhere.
The only cost you need to worry about is the cost per mile that will be about 15 to 20 cents per person for an autonomous ride.
Posted by: Account Deleted | 06 January 2016 at 01:47 PM
This is the equivalent to putting a super computer in the car 30 years ago.
Posted by: SJC | 07 January 2016 at 05:39 AM
SJC the first teraflops supercomputer came to life in 1997. It was a million watt monster occupying a hole house. This shoebox probably uses less than 100 watt and is 25 time faster.
Posted by: Account Deleted | 07 January 2016 at 08:19 AM
You are right, I worked in high tech and image processing for decades.
Posted by: SJC | 07 January 2016 at 09:12 AM
Most of the computing power required will soon be available from improved smart phones or phablets or laptops. Their large touch displays could also replace the car dashboard display.
Low cost redundant electronic hardware availability will not be the main problem for autonomous drive vehicles. Virus proof effective software, more efficient sensors and couplers will need further development.
Human acceptance may be harder to manage.
Posted by: HarveyD | 09 January 2016 at 04:07 PM