Share This article
It’s possible to add self-driving features to future cars for, oh, $150. Seriously. Scientists at Oxford University in England have built a self-driving car around a front-mounted laser scanner. When it sees a roadway it’s capable of navigating, it alerts the driver, who can press a button to give control to the car. Compared to go-anywhere Google self-driving cars, the self-driving car from Oxford dons the role only after comparing the current road to a precise, on-board 3D map of roads it knows about. No GPS is involved. A team at Oxford added self-driving capabilities to a Nissan Leaf.
The all-electric Leaf uses a front-mounted laser scanner, no GPS. In comparison, Google self-driving cars employ a roof-mounted, 360-degree scanner and GPS to verify location. So far, the Leaf has been tested on roads around Oxford at speeds up 40 mph and can stop for pedestrians. According to The Guardian newspaper, “Rather than using the GPS navigation system, which can be unreliable in cities where ‘urban canyons’ caused by buildings block signals, and only accurate to a few metres, the British-developed system uses 3D laser scanning allied to computer storage to build up a map of its surroundings — which is accurate to a few centimetres.” The self-driving system kicks in when it recognizes where the car is, by comparing its surroundings with stored, on-board 3D maps.
The project is led by professor Paul Newman, head of Oxford’s Mobile
Robotics Research Group in the Department of Engineering Science. One of the group’s interests is making vehicles and robots get around when GPS can’t help. “At present many autonomous robots rely on pre-produced maps or GPS to find their way around,” Newman’s bio notes, “but GPS isn’t available indoors, near tall buildings, under foliage, underwater, underground, or on other planets such as Mars — all places we might want a robot to operate.” The research group’s chief achievement seems to be the creation of the FABMAP algorithm, which uses a combination of machine learning and probabilistic inference to compare the current view of a scene with every scene stored in memory.According to Newman, installing the hardware and software on the Nissan Leaf costs about £5,000 (USD $7,620) on top of the cost of the car, which is dirt cheap for a self-driving research vehicle. He envisions self-driving cars being viable in 15 years and the cost could be as little as £100 ($152). As cars change from mechanical controls to drive-by-wire (steering, brakes, throttle), existing car designs would be cheaper to adapt.
A colleague, Martin Spring of Lancaster University, told The Guardian that self-driving cars could change transportation in a big way: The car could be a room on wheels where occupants do what they want while under way; no need for a designated driver. Headlamps wouldn’t be necessary, perhaps not even streetlights. (Unless you wanted to give jaywalkers a fighting chance.)
The Google Lite of self-drivers
The Oxford concept, then, takes a more cost-conscious route to self-driving. It wouldn’t work all the time, but it could work for the bulk of long drives, especially limited-access highways. At the same time, the scanner and 3D maps go far beyond what are likely to be the first so-called self-driving cars that would stitch together existing technologies: adaptive cruise control, lane keep assist, and blind spot detection. That would be enough to self-drive on limited access roads but it would still need a hands-on driver to take back the wheel if a deer or pedestrian darted onto the highway, or if a box fell off a truck. And it’s a reminder that we have to work out liability issues.
Now read: Google: Self-driving cars in 3-5 years. Feds: Not so fast.
Tagged In
Share This Article
via extremetech.com