Lidar is used for sensing obstacles, especially for city driving. It can’t see lane lines, that’s always done with cameras.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Lidar is used for sensing obstacles, especially for city driving. It can’t see lane lines, that’s always done with cameras.
jerjozwik
Active Member
i turned on my 3's autopilot last night for about 15 miles, first thing i noticed was the "hold the steering wheel" notification is VERY obvious. i really like how they show it in the 3, hope that big blue gradient can make it into the S. the current little flag at the bottom of the screen can sometimes be missed.
It not bad, but it’s definitely more apparent in night mode, at least to me. Blue on white doesn’t grab the attenton.
You still need cameras and radar... but Lidar is a major component.
With Lidar you get a 3D cartography of the environment.
The Josh Brown accident would never have happened with Lidar. Lidar would have immediately picked up a large object at the precise location and stopped the car. It doesn't even need to recognize what the freaking object is unlike the cameras which heavily depend on image recognition.
Google would never, ever leave their autonomous cars on the road without Lidar.
We shall see. I’ve done some work with machine learning and vision and I am not convinced. The Josh Brown type incident has also been addressed by altering how the radar works.
Yes, wish they had a dedicated stalk.
Am I the only one who routinely changes the follow distance setting? They buried the follow distance into the autopilot settings requiring multiple touches to adjust from say "5" to "2". Same thing with the speed setting, but at least it's visible and adjustable without going into menus. They should put both of those adjustments right on the main screen.
Ideally, I hope they make the scroll balls on the steering wheel configurable. As of today, the right scroll ball is used for voice commands, ugh..
Sorry but presuming that changing the detection method somehow improves the recognition is extremely naive.
Detection != recognition
Remember that AP1 still had radar which detected the obstacle so detection was never an issue.
Recognizing the obstacle as a semi trailer and not an overhead sign gantry or bridge was/is the real problem - which has nothing to do radar/camera/lidar.
As Tesla have pointed out countless times - humans manage to drive (mostly) without running into things with nothing more than the equivalent of a pair of cameras (eyes). No Lidar or Radar. Just an incredible processing engine behind it.
T-Will
Member
A 3D depth image can be generated using only cameras (LIDAR not required):
The non Lidar approach:
Radar detects something.. radar is low resolution and has a ton of false positives. Camera is then used for vetting using image recognition... Image recognition! Do you think something like image recognition is 100 percent bulletproof?
Lidar approach: object at coordonate X,Y,Z. Boom. Simple and straightforward.
You realize that the LIDAR would output a point cloud that still needs to be processed and analyzed, right?
Not if everybody carries active transponder to convey the information.
Similar threads
