AI Part 4: AI and Self-Driving Cars

Nowadays, Self-Driving Cars has been a very hot topic after Google, Uber and Tesla revealed their plan of achieving Driver-less Cars. In order for the machines to learn driving, the developers added Artificial Intelligence into the self-driving program. 

How is Artificial Intelligence learning how to drive and improve on it?

 

First of all, Deep Learning is used. Sequences are inputted, processed, learned and divided into 4 options: Left, Right, Accelerate and decelerate (brake). 

Important figures such as signs and pedestrians are learned.

For example, the sensor detects something octagon, 

  • 1: Detect if it's red
    • 2: Yes. Any text?
      • 3: Yes. Does it say STOP?
        • 4: Yes, then it is an stop sign, therefore STOP
        • No, keep going
      • No, keep going
    • No, keep going
  • Keep detecting

The same thing also applies to other figures.

Secondly, we put the machine on a simulation program in simple scenes that has no complicated motions, distractions or other vehicles. 

After that, we must add  other variables into the testing, but this time, direct learning doesn't work (simply feeding in sequences).

WHAT NOW?

TO BE CONTINUED . . .

http://www.princeton.edu/~alai...4/Deep%20Driving.pdf

 

PS: The presentation of the Princeton University student helped a lot.

Original Post

 Hi Justin! 

 This is very interesting to read. It's very helpful to know what the process is for AI vehicles and if it will actually work or not. Something to definitely think about in terms of the future of AI is the safety factors. For example, in a car accident, who would the insurance companies blame if the cars were driving themselves. And how would they respond to emergencies?

 Neat research! 

Add Reply

×
×
×
×