Nissan Leaf Prototype Becomes First Autonomous Car On Japanese Highways 140
cartechboy writes "As car manufacturers battle over futuristic announcements of when autonomous cars will (allegedly) be sold, they are also starting to more seriously put self-driving technology to the test. Earlier this week several Japanese dignitaries drove — make that rode along — as an autonomous Nissan Leaf prototype completed its first public highway test near Tokyo. The Nissan Leaf electric car successfully negotiated a section of the Sagami Expressway southwest of Tokyo, with a local Governor and Nissan Vice Chairman Toshiyuki Shiga onboard. The test drive reached speeds of 50 mph and took place entirely automatically, though it was carried out with the cooperation of local authorities, who no doubt cleared traffic to make the test a little easier. Nissan has already stated its intent to offer a fully autonomous car for sale by 2020."
Automotive networks have zero security (Score:3, Interesting)
Car's CAN Bus is ring network with no authentication whatsoever and rudimentary priority system. If you can broadcast into it, then you can affect operation of the car in very drastic ways. Since it has to be real-time and responsive (e.g. controlling engine timing) there is no time for any kind of authentication. Insanity is allowing things like Entertainment/Navigation/OnStar system access to it, but this is how auto engineers do it. Why? Because they don't know any better, they are not IT Security guys.
Re:But does it ... (Score:5, Interesting)
As a side effect, this will finally, finally, FINALLY put an end to the dreaded find-a-parking-space-in-a-busy-city-on-Friday-night drill.
Self-driving cars can not only use remote parking lots, they can also make much better use of parking lot space. They are unoccupied when they self-park, so there is no need to leave room for people to exit. So they can park just an inch apart, and the absence of side mirrors will make that very close. Less space is needed for lanes, since the cars can steer optimally and coordinate their movements. Cars could park directly in front and behind each other, then when summoned by its owner, a car could signal for the blocking cars to move. The capacity of a parking lot can easily be doubled or tripled.
Re:Glitch = Possible Death (Score:4, Interesting)
What if it suddenly veers into a wall or oncoming truck due to an incorrect or faulty instruction. Fuck autonomous!
You are obviously not an embedded system engineer with mission critical design experience. The proper way to design a system like this is to have multiple processes running on at least two separate CPUs. The most powerful CPU computes the car's speed and path, and another process running on a separate CPU performs sanity checks on the results. If something is clearly wrong (like steering into oncoming traffic), then the backup program applies the brakes and pulls off the road. Bits can be flipped by cosmic rays, or whatever, and a system like this is designed to deal with that. This is standard critical system engineering. Then you put it on the test track, and throw all the crap you can at it: turn off sensors at random, put corrupt data on the bus, flip bits in memory, etc. Keep hammering it and fixing the problems until it can handle any failure as safely as possible.