3 ms·

As a Canadian, it's the snow that's made me skeptical of self-driving cars. Seeing how much they depend on road markings to navigate in traffic properly is comp

As a Canadian, it's the snow that's made me skeptical of self-driving cars. Seeing how much they depend on road markings to navigate in traffic properly is completely impossible in winter. Until that's solved, there's a lot of places that just can't use them for a lot of the year.

cal5k 8y ago
It seems plausible to me that LIDAR + radar could eventually outperform humans in white-out conditions. Here's an interesting bit of research on filtering out snow noise from a LIDAR point cloud: http://wavelab.uwaterloo.ca/?weblizar_portfolio=real-time-filtering-of-snow-from-lidar-point-clouds http://wavelab.uwaterloo.ca/?weblizar_portfolio=real-time-fi...
- tlear 8y ago
  In those conditions maybe, problem is where is the road.. or lane. There are no markings of any kind when you have snow on the road. No marking you need to estimate where lanes are from what is near road and what other cars are doing. Exponentially harder problem
- gugagore 8y ago
  But if snow occludes lane markings, which is what the parent was discussing, then no amount of sensing helps you see them, right? (You can have a map with lane markings and localize on that map, though)
  - jpindar 8y ago
    Except that many times when the lines are invisible, new lanes form that are in a different location. So your map will not match up with what the rest of the traffic is doing.
    - gugagore 8y ago
      I agree. I was just anticipating the common response to that challenge, or any other "static" perception challenge, which is "but there is a map"
      - cal5k 8y ago
        I mean... the map at least bounds the problem. You can be pretty certain in almost every case that the lanes won't be outside of the mapped roadway.
wongarsu 8y ago
Waymo seems much less focused on road markings than Tesla and Uber, and instead more focused on the geometry problem of not hitting stuff. They are also in a great position to analyze summertime satellite images to get a better understanding of road markings. I'm optimistic that this is a very solvable problem for Waymo.
- Piskvorrr 8y ago
  Solvable, indeed - in the Dartmouth-1956-workshop meaning of "solvable": there's as of yet an unknown amount of research to be done, and it looks reasonably complex to be hacked at by a dozen people over the summer. Therefore, a few man-centuries of research and development will surely bear fruit.

cal5k 8y ago

It seems plausible to me that LIDAR + radar could eventually outperform humans in white-out conditions. Here's an interesting bit of research on filtering out snow noise from a LIDAR point cloud: http://wavelab.uwaterloo.ca/?weblizar_portfolio=real-time-filtering-of-snow-from-lidar-point-clouds http://wavelab.uwaterloo.ca/?weblizar_portfolio=real-time-fi...

tlear 8y ago
In those conditions maybe, problem is where is the road.. or lane. There are no markings of any kind when you have snow on the road. No marking you need to estimate where lanes are from what is near road and what other cars are doing. Exponentially harder problem
gugagore 8y ago
But if snow occludes lane markings, which is what the parent was discussing, then no amount of sensing helps you see them, right? (You can have a map with lane markings and localize on that map, though)
- jpindar 8y ago
  Except that many times when the lines are invisible, new lanes form that are in a different location. So your map will not match up with what the rest of the traffic is doing.
  - gugagore 8y ago
    I agree. I was just anticipating the common response to that challenge, or any other "static" perception challenge, which is "but there is a map"
    - cal5k 8y ago
      I mean... the map at least bounds the problem. You can be pretty certain in almost every case that the lanes won't be outside of the mapped roadway.

wongarsu 8y ago

Waymo seems much less focused on road markings than Tesla and Uber, and instead more focused on the geometry problem of not hitting stuff. They are also in a great position to analyze summertime satellite images to get a better understanding of road markings. I'm optimistic that this is a very solvable problem for Waymo.

Piskvorrr 8y ago
Solvable, indeed - in the Dartmouth-1956-workshop meaning of "solvable": there's as of yet an unknown amount of research to be done, and it looks reasonably complex to be hacked at by a dozen people over the summer. Therefore, a few man-centuries of research and development will surely bear fruit.