Am I still right that all self driving cars still make decisions using old school control theory, and AI is still used only for sensors, and there's no plans to change this?
Conversation
13
There is absolutely no reason to change that
Quote Tweet
Real question: What will a self driving car actually do - swerve to a lane that it has less data about or keep course and brake hard? I would keep the advantage of data because who knows if there is a cliff?
False question here reveals more about ethics of people than cars. twitter.com/wef/status/105…
Show this thread
1
3
Show replies
1
1
Show replies
No. People are certainly looking at neural net trained control algorithms.
I would assume for safety critical applications there would be some old school limit parameters etc., but not really my field so can't tell.
1
3
I'm a bit more familiar with what happens on the air than on the ground, but, it's complicated. I don't see control theory going away any time soon,
5
but we do see a growing interest in using it within a simplex architecture where the classic controls only kick in when an ml algorithm tries to do something that lies outside the envelope of safe actions
5
This gives you the flexibility to optimize for secondary metrics like comfort or efficiency without sacrificing provable end to end safety
3
Source: it's my PhD dissertation, albeit my work is the proofs side and just leaves a hole open where the ML can go
5







