You’re driving along in the right lane of a busy four-lane street, minding your speed and obeying all the rules of the road. Up ahead you spot an obstacle—two sandbags surrounding a storm drain. You slow down and check the traffic in the lane to your left, where you see a city bus lumbering along, some distance behind you but gradually getting closer. Now you have a choice: swing into the left lane in front of the oncoming bus or continue in your present lane and run over a sandbag.
Being an experienced driver, you choose the second option. You know that the bus will have a hard time slowing down, so cutting in front of it is inviting disaster. You also know that bumping over a sandbag or two won’t damage your car.
Unfortunately, when a self-driving car encountered such a situation this past February, it chose the first option. The car pulled into the bus' lane, with the predictable fender-bender outcome.
Unfortunate result of self-driving car's encounter with a bus
Could a Kinect sensor have helped the autonomous car make the better decision? Researchers at the University of La Laguna in the Canary Islands think so. They are building a system that uses the Kinect sensor’s infrared depth camera to help self-driving cars identify nearby obstacles and maneuver around or over them.
Using a self-driving golf cart, the researchers tested the Kinect depth camera against a laser rangefinder and stereo cameras. By operating the vehicle on a test road that contained stairs, ramps, and curbs, they discovered that the Kinect sensor outperformed the other two devices in detecting and correctly discerning nearby, close-to-the-ground objects. For instance, when confronted by a ramp, the laser rangefinder mistakenly determined that it was too steep to drive up. The laser device also failed to spot lower stairs, while the stereo camera had trouble identifying very close (and very far) objects and gave frequent false detections.
The Kinect sensor correctly identified the ramp as navigable and consistently outperformed the stereo cameras in detecting obstacles close to the ground. Lead researcher Javier Hernandez-Aceituno praised the Kinect sensor’s superior abilities in detecting close-range obstacles, telling reporters that the Kinect sensor allows an autonomous vehicle to navigate safely in areas where the other systems fail. And isn’t that reassuring to both future passengers and pedestrians?
The Kinect for Windows Team