[mygoodaccount]

Slightly off topic, but it's fun to see company culture reflected in the algorithm behaviour.

There are thousands of clips of Tesla's Fatal Self Driving being far too aggressive for American roads (https://nitter.net/taylorogan/status/1469404579439824899).

Meanwhile, companies like AutoX have the problem of being far too conservative for Asian roads (https://www.youtube.com/watch?app=desktop&v=TFEvkmvIjVo). AutoX has some of the best edge case handling I've seen.

[intrasight]

This I've long thought the be the Achilles' heel of autonomous. Even skilled human drivers have a hard time calibrating their style to the local situation.

I remember the first time I drove as a passenger through Manhattan with my father driving, and couldn't get my head around his change in driving style. Was a side of him I'd never seen before - driving aggressively like a cab driver. I'm sure it came from his years of living in NYC.

[wpietri]

For sure. Driving on a test track is a technical problem. Driving in traffic is a social problem. As you noted, how one drives in traffic is not just about getting somewhere, but about communicating and negotiating with other drivers around you.

It's one of the things that makes me think that true self-driving may be in the AGI bucket.

[capnhawkbill]

If all the cars would be autonomous then this problem would be solved because they would all keep the same driving style right?

[wpietri]

Not at all. You really think somebody wouldn't program their cars to be more aggressive so that their owners get there faster?

[intrasight]

I think there will have to be a NYC mode

[ec109685]

Learning the appropriate driving style for a neighborhood / city seems like one of the easier AI problems for Tesla to solve.