Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by rvalue | Hacker News Reader

Full thread

rvalue·

I don't think these autopilot systems will work well in crowded cities or cities with poor infrastructure for roads.

The amount of objects for detecting and avoiding will be way too high.

The tests shows almost clear conditions for driving. This should be tested on streets of NY or a busy city like Mumbai

View on HN

> I don't think these autopilot systems will work well in crowded cities or cities with poor infrastructure for roads.

I would _love_ to see some serious "read team" testing of these exact issues with driverless cars. Pedestrian issues, weather issues, mixed-use roads, roads with construction crews, roads with simulated accident scenes, roads with physical damage/flooding/large debris.

I'd also be interested to know if the car drives differently between the rear-wheel only and the all-wheel versions and if environmental factors can frustrate these different control modes.

Check GM autopilot demos in the streets of San Francisco. Much more impressive than this one. Pedestrians, bicycles, delivery van stopped in the middle of the street, etc.

daytime: https://www.youtube.com/watch?v=SSc8AHIEG9o

night time: https://www.youtube.com/watch?v=XHIq0MMViZg

Cruise is also using Lidar which provides depth sensing[1]. You can do depth sensing with cameras, but that requires building depth maps from multiple cameras (like eyes) and still being able to interpret the depth map with computer vision. Capturing depth is definitely possible with monocular cameras (because humans can do it), but it seems like lidar works a lot more accurately.

[1] - https://www.getcruise.com

Also, why isn't there a test for night time ?

What a controversial opinion /s

Reply on news.ycombinator.com