Deep-learning supercomputers bring autonomous cars closer to reality
The big carmakers already have most of what they need: adaptive cruise control, lane-keeping technology, collision avoidance and automatic lane changing. Armed with video cameras, radar and lidar (laser) sensing equipment, the automobile of the future is morphing into a Âpeculiar beast.
And there’s new technology to make it smarter. At this month’s CES technology show in Las Vegas, chipmaker Nvidia launched the Drive PX 2 in-Âvehicle artificial intelligence platform for driverless operation.
Nvidia says the PX2 has the processing power of 150 MacBook Pros, and can perform 24 trillion deep learning operations per Âsecond.
“It utilizes deep learning on Nvidia’s most advanced GPUs for 360-degree situational awareness around the car, to determine Âprecisely where the car is and to compute a safe, comfortable trajectory," explained chief executive Jen-Hsun Huang.
Deep learning involves supercomputers in the cloud capturing all the unexpected random events that occur on the road — swerving cars, bikes meandering across lanes, sudden rainstorm and road works. The learning from data collected can be shared with all vehicles using the system.
The PX2 also offers sensor Âfusion, localization and path planning. Sensor fusion involves receiving inputs from up to 12 cameras, radar, ultrasonic sensors, and lidar, and then fusing it to develop a comprehensive picture of where the car is, the traffic conditions it needs to contend with and plot path scenarios for the car to travel down lanes.
The Drive PX 2 development engine will be available in the fourth quarter of this year and Nvidia says Volvo will use it in the SUVs deployed at its driverless trials next year.
We won’t be taking our hands off the wheel until manufacturers and the likes of Nvidia demonstrate that the technology is watertight. Lots of redundancy needs to be built into autonomous systems: what happens if the computer chip or sensors fail? There are also serious issues around liability and insurance that will need to be ironed out before Âdriverless cars and the more traditional automobiles are ready to share the road.
Meanwhile, the public will need to be convinced that the technology is safe as governments and regulators look to sign off on its use. But there are powerful forces at work that want this to happen sooner than later. In Australia, the South Australian government is keen to declare sections of highways driverless-enabled and there’s interest in other states. Tollway firms want to offer a driverless experience. Last year, Transurban conducted driverless vehicle trials in northern Virginia in the US. And in Nevada the law allows for the testing and operation of driverless cars, although only testing is permitted at this time.
When the driverless era begins, we are likely to see the autonomous experience rolled out bit by bit, first on controlled sections of divided roadways where traffic flow is orderly and there is the least likelihood of unpredicted obstructions. You will drive normally to those controlled sections, take your hands off the wheel while on them, and then resume control as you prepare to leave.
If Tesla has its way, driverless cars will be on the road before 2020. Founder Elon Musk wants you to summon a driverless Tesla from across country and Tesla plans to have cars using Autopilot software ready by 2018. It’s currently rolling out a software update that lets some models park themselves provided they are within about 10 meters of a parking space. The driver has to be within three meters of the car. A summon function will let drivers retrieve their Tesla nearby.
Driverless parking is already a reality. At CES, BMW showed off the i3’s ability to park itself without anyone aboard. A driver controls the process from afar and later summons the car from the spot before getting in. BMW also displayed its Vision Future concept vehicle. It includes an option to angle the front seats towards one another to aid conversation. After all, why should the seats perpetually face forward when the driver is no longer in control?
It’s a sign of things to come and the carmakers are firing on all cylinders to win pole position. As The Australian’s motoring editor Philip King has reported, the new Mercedes E-Class unveiled at the Detroit Motor Show could drive itself for up to a minute. Capabilities include autonomous overtaking and steering through inner-city traffic without the need for lane markings. Mercedes’ research department has spent years developing the artificial intelligence needed for autonomous cars and most of the self-drive features in the E-Class, except remote-control parking, will be fitted on cars in Australia.
At CES, Toyota hosted a futuristic display of co-ordinated driving and artificial intelligence. A group of model cars drove autonomously around a small area, sharing their movements with the cloud. At first, the cars frequently collided, but over time learnt each other’s movements. Toyota wants driverless cars on the road by 2020.
Meanwhile, some initiatives are focused on connected technology. Apple CarPlay and Android Auto entertainment systems will be universal on all SYNC-equipped Ford vehicles from 2017 in North America, with the features expanding worldwide after that. Driverless tech development is also being backed by large investments, such as the $US500 million ($724m) General Motors is investing in ride-sharing firm Lyft. The long-term aim for GM is to develop a network of on-demand autonomous cars, where you would ring up and summon a vehicle when needed.
It’s a brave new world of automobiles but the questions remain: can you trust a computer to see all objects on an open road, and for it not to fail? When it’s working, a bevy of cameras, radar and sensors might see much more than two human eyes.
Computers don’t drink-drive and don’t text while driving. But it will take time for humanity to trust machines hurtling down the road with no human behind the wheel. BusinessSpectator.com.au