Discovering the technologies behind the autonomous vehicle
The concept of autonomous vehicles is driving dynamism and innovation in the components industry. Sensors, power converters, GPS systems, wireless modules, control and communications technologies, and HMI products are all being developed with the autonomous vehicle in mind. In the meantime, Advanced Driver Assistance Systems (ADAS) are already producing revenue for the component makers. Here’s a peek at some of the most interesting developments announced in the last few months.
Neural network-based driving platforms
The first generation of self-driving cars used LIDAR systems to recognize terrain and obstacles. While accurate, LIDAR technology is bulky and expensive, making it hard to integrate into mass produced consumer vehicles where price and design matter. That’s why the newest generation of autonomous vehicles is centered around computer vision. Rather than using expensive, specialized LIDAR systems, simple cameras are used in conjunction with sophisticated machine vision techniques.
While the hardware is simple, the software and processing involved is anything but. The current state-of-the-art computer vision techniques for autonomous driving involves deep learning using convolutional neural networks (CNN). Rather than having humans describe objects – lanes, curbs, traffic signs – which the system tries to detect, CNN based deep learning systems analyze vast quantities of images and recognizes patterns of objects. Besides being efficient, CNN-based computer vision is robust, able to handle situations such as inadequate lane markings as well as poor visibility situations caused by inclement weather or poor roadside lighting.
Vehicle to X (V2X) coming to market
[Figure 1 | The 2017 Cadillac CTS will feature V2X communications using DSRC technology. (Source: GM)]
Self-driving technology is improving every day, but as crashes by self-driving Google cars and Tesla models on auto-pilot mode demonstrate, camera and even LIDAR-based approaches may not be sufficient to prevent all crashes. A number of Google car crashes have occurred when the vehicle itself followed the rules of the road, but human drivers around it misbehaved. Another crash involving a Google car occurred when the self-driving vehicle misjudged the intentions of a bus traveling behind it and crashed into the side fender.
Tesla’s self-driving record hasn’t been spotless either. In June of this year, a Tesla owner was killed when the car, driving in auto-pilot mode, crashed into a tractor trailer turning in front of it. It’s suspected that the computer vision system failed to detect the white truck body against the background of the bright sky that day.
What these crashes have in common is that they could all have been prevented if the cars involved had V2X technology.
V2X stands for Vehicle to X, where X could be other cars, buildings, traffic lights, or even pedestrians. Using short-range radio technology, V2X-capable vehicles communicate information with each other such as vehicle speed and direction of travel to help avoid collisions.
The radio technology behind V2X is called dedicated short range communications (DSRC), designed specifically for car-to-car communications. DSRC is based on the 802.11 Wi-Fi standard, but is optimized for extremely low latency, ad-hoc connections, allowing cars to communicate quickly while traveling at high speeds to help avoid collisions.
With V2X communications, cars have a much better, more precise picture of the speed and position of vehicles in the surrounding area, even beyond line of sight. Information on surrounding cars can be picked up around blind corners or in low-visibility situations such as at night or in dense fog that would baffle camera-based technology.
A long time coming, the technology’s main obstacle has been a chicken-and-the-egg problem, as V2X is most effective when many cars on the road have it. Thankfully, it is finally making its way to the market, with the 2017 Cadillac CTS and 2017 Mercedes E-Class featuring V2X communications to help ward off accidents. Along with more and more DSRC chips becoming available on the market, this nascent, potentially life-saving technology may soon become widespread.
The self-driving car kit
[Figure 2 | Comma.ai's self-driving kit promises to integrate into cars such as the 2017 Acura ILX which have drive-by-wire capability. (Source: Acura)]
Perhaps the most exciting recent news for self-driving car enthusiasts is the upcoming availability of a sub-thousand-dollar autonomous driving kit from Comma.ai. Created by George Hotz, famous for hacking the iPhone, this kit will be a $999 retrofit to add self-driving capability to cars which have electronic stability control and electronic power steering. While the price is cut-rate, the software is not. Comma.ai’s system uses state of the art convolutional neural networks for its computer vision.
The kits are expected to ship by the end of this year. In the meantime, Comma.ai is recruiting drivers to help improve its neural network by downloading its Dash app. Users mount their smartphone on their dashboard to record actual driving data, which is used to train its neural network. With a properly trained network, Comma.ai promises its kits will rival Tesla’s Auto-Pilot Mode.
Only time will tell whether the kits will work (or ship) as expected, but one thing is for certain: the future of autonomous driving looks bright, and it’s coming much faster than any of us would have expected.