Nissan's Crash-Free R&D: 7 Cute Robots Mimicking Bees and Fish 105
cartechboy writes "As Nissan develops autonomous cars for its 2020 target date, the company's engineers are modeling the tech after behaviors seen in bumblebees and fish. Nissan actually tests self-navigation algorithms in seven small toy-looking robots called EPORO. The robots have 180-degree vision (modeled after bees) and monitor each others' positions, travel nose to nose and avoid collisions--just like a school of fish. Getting small robots to zip around without bumping into things might be the first step in getting cars to do the same."
Re:Seems a stretch (Score:4, Insightful)
not all cars have to be autonomous for this to work. The autonomous cars will just be traveling in groups, which will be very efficient. I can even foresee dedicated lanes for autonomous cars.
Re:Why? What fun is an autonomous car? (Score:2, Insightful)
Every driven I-10 across Texas? Try it and you'll understand why you might want an autonomous car.
Re:a wet blanket (Score:4, Insightful)
If you want to make a human comparison, you have to ask, "how often does a human have a heart attack or other catastrophic failure on the road?" This is just the base system, you need to make it reliable before you even get to talking about the quality of the algorithms. It's great you have a perfect driving algorithm, but if the OS crashes and needs to reboot once a year while driving, then no one will care if you have a perfect algorithm. Your car is having the equivalent of a heart-attack every year on the road.
It's harder than writing a typical website. To get an idea of the difficulties involved, remember that at that level of reliability you need to take into consideration that cosmic rays will corrupt your memory. ECC can be helpful, but you can still get corruption. So imagine if you have a for loop, and your index gets corrupted. How will you guarantee that it doesn't become an infinite loop?
These problems are known and can be solved, but the point is, it's really hard. It will take six years of effort, and that is assuming the algorithms are ready now. If they are not, then they won't have it ready by 2020.