NVIDIA Unveils Palm-Sized, Energy-Efficient AI Computer for Self-Driving Cars

[btarunr]

NVIDIA today unveiled a palm-sized, energy-efficient artificial intelligence (AI) computer that automakers can use to power automated and autonomous vehicles for driving and mapping. The new single-processor configuration of the NVIDIA DRIVE PX 2 AI computing platform for AutoCruise functions -- which include highway automated driving and HD mapping -- consumes just 10 watts of power and enables vehicles to use deep neural networks to process data from multiple cameras and sensors. It will be deployed by China's Baidu as the in-vehicle car computer for its self-driving cloud-to-car system.

DRIVE PX 2 enables automakers and their tier 1 suppliers to accelerate production of automated and autonomous vehicles. A car using the small form-factor DRIVE PX 2 for AutoCruise can understand in real time what is happening around it, precisely locate itself on an HD map and plan a safe path forward. "Bringing an AI computer to the car in a small, efficient form factor is the goal of many automakers," said Rob Csongor, vice president and general manager of Automotive at NVIDIA. "NVIDIA DRIVE PX 2 in the car solves this challenge for our OEM and tier 1 partners, and complements our data center solution for mapping and training."



More than 80 automakers, tier 1 suppliers, startups and research institutions developing autonomous vehicle solutions are using DRIVE PX. DRIVE PX 2's architecture scales from a single mobile processor configuration, to a combination of two mobile processors and two discrete GPUs, to multiple DRIVE PX 2s. This enables automakers and tier 1s to move from development into production for a wide range of self-driving solutions -- from AutoCruise for the highway, to AutoChauffeur for point to point travel, to a fully autonomous vehicle.

The new small form-factor DRIVE PX 2 will be the AI engine of the Baidu self-driving car. Last week at Baidu World, in Beijing, NVIDIA and Baidu announced a partnership to deliver a self-driving cloud-to-car system for Chinese automakers, as well as global brands.

"Baidu and NVIDIA are leveraging our AI skills together to create a cloud-to-car system for self-driving," said Liu Jun, vice president of Baidu. "The new, small form-factor DRIVE PX 2 will be used in Baidu's HD map-based self-driving solution for car manufacturers."

NVIDIA DRIVE PX is part of a broad family of NVIDIA AI computing solutions. Data scientists who train their deep neural networks in the data center on the NVIDIA DGX-1 can then seamlessly run on NVIDIA DRIVE PX 2 inside the vehicle. The same NVIDIA DriveWorks algorithms, libraries and tools that run in the data center also run in the car.

This end-to-end approach leverages NVIDIA's unified AI architecture, and enables cars to receive over-the-air updates to add new features and capabilities throughout the life of a vehicle.

Product Specifications
NVIDIA DRIVE PX 2 is powered by the company's newest system-on-a-chip, featuring a GPU based on the NVIDIA Pascal architecture. A single NVIDIA Parker system-on-chip (SoC) configuration can process inputs from multiple cameras, plus lidar, radar and ultrasonic sensors. It supports automotive inputs/outputs, including ethernet, CAN and Flexray.

Availability
The new single-processor DRIVE PX 2 will be available to production partners in the fourth quarter of 2016. DriveWorks software and the DRIVE PX 2 configuration with two SoCs and two discrete GPUs are available today for developers working on autonomous vehicles.

View at TechPowerUp Main Site

 

[xvi]

Gonna need Wiz to tell me how good it'll be for F@H.

 

[RejZoR]

I always wondered who'll be to blame when such car crashes. The car manufacturer? NVIDIA? Driver owning the car? Who? Because we know they WILL crash at one point. It's just inevatable. No computer system can predict certain things the way human mind can.

 

[Ungari]

Another problem with car accidents are the decisions made by the AI as to what lives will be protected, and who will likely not survive.
Programmers of these machines are already grappling with the moral implications, but machines can dispassionately make valuations on who is worth saving, and who is expendable, when choosing what to do when a crash is imminent and cannot be avoided. Insurance companies and their exposure to liability will almost certainly be a factor in this as well.

 

[ViperXTR]

b-but nvidia always had good drivers :V

 

[the54thvoid]

The AI isn't comparable to our sci if interpretation. It uses mapping and collision avoidance, it doesn't have any opportunity to make decisions on the value of life. I think a comment above is thinking o, or at least similar to 'I, Robot' (the film) where a robot calculates a chance of death versus life over two people.
The hero hates AI for choosing his life over a young girl based on survival chance. AI for driving is not in that position, it simply chooses action to avoid and reduce damage collision.
Yes it will not always work but it will be accidental with limited to zero liability, as it is now with human error. Liability may also be with the human as it would be easy to place the human under acceptance of liability by proxy for using such a system.

 

[Ferrum Master]

That's the most powerful ARM CPU right now. I wonder if it would end up in some other device.

 

[RejZoR]

b-but nvidia always had good drivers :V

AHAHAHAAHAHAHAAHAH

Yeah, their drivers generally don't crash, but you never know. You just opened a can of puns

 

[Xzibit]

b-but nvidia always had good drivers :V

Wondering if a car equipped with one of these will force you to login through facebook before it opens the door for you to get in.

 

[ViperXTR]

nvidia can't async though, so they are gonna be slower, but maybe safer? :V :V

 

[seronx]

Nothing beats NXP's i.MX 8...

 

[ViperXTR]

Wondering if a car equipped with one of these will force you to login through facebook before it opens the door for you to get in.

Only if you have GFE 3.0 installed on your car i hear

 

[Caring1]

Wondering if a car equipped with one of these will force you to login through facebook before it opens the door for you to get in.

The built in entertainment center will most likely have Skype as a default program too.

 

[$ReaPeR$]

I always wondered who'll be to blame when such car crashes. The car manufacturer? NVIDIA? Driver owning the car? Who? Because we know they WILL crash at one point. It's just inevatable. No computer system can predict certain things the way human mind can.

that's the big question my friend! legislation has and is quite behind the technological progress of our societies and the only way to fix that is by having knowledgeable people create such legislation, but no one seems to be interested, so we are doomed to face the concequences.

 

[RejZoR]

They'll also need to have an event log when driver is using auto pilot mode and manual mode. Because if the car crashes while manual driving, one could easily blame auto pilot for it. Then again company like NVIDIA could blame it on the driver. Not sure if one would go that far though.

 

[Caring1]

Easy solution, just install Windows 10, then the driver will have no control

 

[ZoneDymo]

Install generic windows drivers, see what happens

 

[Ungari]

The AI isn't comparable to our sci if interpretation. It uses mapping and collision avoidance, it doesn't have any opportunity to make decisions on the value of life. I think a comment above is thinking o, or at least similar to 'I, Robot' (the film) where a robot calculates a chance of death versus life over two people.
The hero hates AI for choosing his life over a young girl based on survival chance. AI for driving is not in that position, it simply chooses action to avoid and reduce damage collision.
Yes it will not always work but it will be accidental with limited to zero liability, as it is now with human error. Liability may also be with the human as it would be easy to place the human under acceptance of liability by proxy for using such a system.

There was an article discussing the moral hazards programmers faced while working on this technology, and so yes the AI are being programmed make a decision on who lives and who dies in certain scenarios.
It is really no accident that you have already seen this portrayed in fiction, as SciFi books and movies like I, Robot often externalize to the public what is planned for the future. So many things that were considered far out a few years ago, are now a reality. Think George Orwell's 1984 written in 1948, where Flat-screen TVs that watch you are now yesterday's news.

 

[KainXS]

I would like to know if this drive system can detect when a driver is Incapacitated, has a Myocardial event or a Seizure of some kind as if it can and it could really save some lives. If it could detect that the driver is not longer capable to drive it could begin driving to the nearest hospital and phone 911 in the process. It could also take over before a collision and pull over to the shoulder of the road to prevent accidents.

When it comes to the ethical choices around loss of life I think that no matter what should a driver be allowed to remove their hands from the steering wheel if only for a second. If there comes to be a choice around a fatal accident and the driver can take over in time they should be ready. There could however be a discussion about if the machine could have done a a better job in such a situation as a human would not be able to react faster. There needs to be moral programming if machines will make these choices though, where a human dies similar to Isaac Asimov's(his rules might be too rigid though) but when your trading one life for another his first rule and 0th rule is violated no matter what. I am pretty sure that the drive system already has such rules, and googles, ubers, and whoever is doing such an AI project also has them.

 

[Frick]

I always wondered who'll be to blame when such car crashes. The car manufacturer? NVIDIA? Driver owning the car? Who? Because we know they WILL crash at one point. It's just inevatable. No computer system can predict certain things the way human mind can.

Me I wonder about longevity. I currenty drive a 1991 Volvo 740. All these electric cars and the systems they have, I'm pretty sure they won't last that long.

 

[Ungari]

I would like to know if this drive system can detect when a driver is Incapacitated, has a Myocardial event or a Seizure of some kind as if it can and it could really save some lives. If it could detect that the driver is not longer capable to drive it could begin driving to the nearest hospital and phone 911 in the process. It could also take over before a collision and pull over to the shoulder of the road to prevent accidents.

When it comes to the ethical choices around loss of life I think that no matter what should a driver be allowed to remove their hands from the steering wheel if only for a second. If there comes to be a choice around a fatal accident and the driver can take over in time they should be ready. There could however be a discussion about if the machine could have done a a better job in such a situation as a human would not be able to react faster. There needs to be moral programming if machines will make these choices though, where a human dies similar to Isaac Asimov's(his rules might be too rigid though) but when your trading one life for another his first rule and 0th rule is violated no matter what. I am pretty sure that the drive system already has such rules, and googles, ubers, and whoever is doing such an AI project also has them.

Here is an example of a scenario where the AI would have to weigh the probability of casualties and make a choice based on what programming it has been given:

Two of these self-driving vehicles are traveling down a street when the lead car has a malfunction that causes the brakes to suddenly lock-up. The following car knows it cannot stop in time to avoid hitting the lead car so it has three choices:

A) Apply brakes and hit the lead vehicle at reduced speed and hope for the best, the AI knows the occupants of both vehicles are families with children and earn low to average incomes with high healthcare costs paid for by the government.

B) Make a sharp Left Turn and crash into Jen-Hsun Huang who is recognized through facial recognition and is sitting outdoors at at sidewalk cafe.

C) Make a sharp Right Turn and crash into Lisa Su who is trying to cross the street to say hello to Jen-Hsun.

The AI does rapid calculations of all liability costs and outcomes and decides option C is best.

 

[Totally]

I always wondered who'll be to blame when such car crashes. The car manufacturer? NVIDIA? Driver owning the car? Who? Because we know they WILL crash at one point. It's just inevatable. No computer system can predict certain things the way human mind can.

The car manufacturer everyone else is shielded behind them. Any responsibility or contractual agreements are between the car manufacturer and the consumer regardless who in the line is at fault if the product fails since the final say on the finish product ultimately falls to them. It the fault lies with another party involved then it is up to the car manufacturer to pursue it how they see fit. For example, take Samsung and their exploding batteries. Say the batteries were sourced from a third party and you bought a Note 7 one day and that very night battery exploded and burnt your house down. You can't file a suit against the battery manufacturer and have to take it up with Samsung since it was their phone you bought. Then it's Samsung's turn to deal with the battery maker for giving them faulty batteries should it choose to do so.

 

[CAPSLOCKSTUCK]

Can you get drunk and drive an auto pilot car?

 

[Caring1]

Can you get drunk and drive an auto pilot car?

Technically you would be a passenger and not in charge, so you can't be charged for drunk driving.
It shouldn't be any different than stepping on to a self driving bus when inebriated.

 

[GreiverBlade]

AHAHAHAAHAHAHAAHAH

Yeah, their drivers generally don't crash, but you never know. You just opened a can of puns

it's not : "Yeah, their drivers generally don't crash"
it's : "Yeah, their drivers usually crash quit often"

the "geforce driver are better" is proven to be massively wrong (latest 10XX episode was funny )

worst driver i've ever had? driver that crashed the most? most unstable drivers?
well most of the time Forceware over Catalyst and that, since the beginning

(not counting SIS 6326 card and Xabre also ) not that the ATI/AMD Catalyst driver never crashed on me... but it was way rarer and also i never had a unstable OC with ATI/AMD while Nvidia was ... well ... mixed result

i also wonder who you could blame in case of a car crash ... i would like it to be Nvidia ... since the idea originate from them, car manufacturer can also be, driver too ... because the only main source of idiocy and stupidity in our world is, and will always be, human.

computer can predict things better than human do, only if programmed correctly at the start, only human make mistake (yep even computer mistake has a human origin)