4 ms·
I don't know if they have made it into production yet (probably not?) but Lindsey Kuper wrote a nice pair of posts on how DNNs can replace an existing collision
by moyix 5y ago
I don't know if they have made it into production yet (probably not?) but Lindsey Kuper wrote a nice pair of posts on how DNNs can replace an existing collision avoidance system (ACAS), and how to verify whether it's still safe after doing so:
http://composition.al/blog/2017/05/30/proving-that-safety-critical-neural-networks-do-what-theyre-supposed-to-where-we-are-where-were-going-part-1-of-2/ http://composition.al/blog/2017/05/30/proving-that-safety-cr...
http://composition.al/blog/2017/05/31/proving-that-safety-critical-neural-networks-do-what-theyre-supposed-to-where-we-are-where-were-going-part-2-of-2/ http://composition.al/blog/2017/05/31/proving-that-safety-cr...
- light_hue_1 5y agoACAS X is like the sweet spot for AI. We have a system now, TCAS II which is reliable but has a lot of false positives and has a big limitation: both aircraft need to have TCAS in order to detect and resolve a conflict. It's also an environment that is very simple to simulate and model mathematically compared to virtually anything else AI will ever be applied to. TCAS II will be around for a very long time, like two decades, while ACAS X is deployed. So the AI system will also have a backup that already works. This is really the perfect target for AI: clear backups, we just need some extra capabilities and warnings, easy to simulate, false positives are acceptable. That's basically unique. Even then nothing has been deployed yet. We're half a decade away or so at best.