3 ms·

"That is, it’s entirely possible that Waymo, or someone else, gets autonomy to work in 202x with a $1000 or $2000 LIDAR and vision sensor suite and Tesla still

"That is, it’s entirely possible that Waymo, or someone else, gets autonomy to work in 202x with a $1000 or $2000 LIDAR and vision sensor suite and Tesla still doesn’t have it working with vision alone." In this scenario there still may be an upside for Tesla. For Waymo to cover the world in LIDAR-equipped cars will take decades. Tesla already sells cars all over the world. The Tesla car sharing network may scale extremely quickly as a result. Even if Waymo takes some markets first, once Tesla has autonomy working, they can be in most markets at the flip of a switch. Since the cars participating in the sharing network would be from Tesla purchasers, people may purchase the cars as an investment, meaning there would be no capital cost to Tesla in scaling their network. They just need to be able to produce enough Model 3s and get autonomy working. Then they can take profits from both the sale of the car and the clip of the car sharing network.

kbob 8y ago
Is LIDAR actually better than video? If I understand LIDAR correctly, it gives a 1D array of depth data and is fairly limited in either resolution or speed. Video provides much more data at higher speed, and it seems like it'd be easier to identify objects from 2D color images. Stereo (never mind eight cameras) can be used to derive depth information. But I'm not a practitioner, so educate me. How does LIDAR beat video?
- jfoster 8y ago
  I was quoting the article, but I think LIDAR is a bit more like having a 3D model of the world around the vehicle. The processing of a 1D array into a usable point cloud is not of concern at the frontier of autonomous vehicle research. My understanding is that the main reason Tesla would choose video instead is that cameras are super-cheap in comparison to LIDAR, and theoretically good enough if you have AI that can understand what the cameras are seeing. I have a feeling that Musk is pushing OpenAI toward researching the technology that Tesla will need to make their approach work. I hope they are successful; it is probably the fastest path toward ubiquitous autonomous vehicles. It's likely that Waymo can get theirs working faster, but they can't scale the distribution side as fast as Tesla could.

kbob 8y ago

Is LIDAR actually better than video? If I understand LIDAR correctly, it gives a 1D array of depth data and is fairly limited in either resolution or speed. Video provides much more data at higher speed, and it seems like it'd be easier to identify objects from 2D color images. Stereo (never mind eight cameras) can be used to derive depth information. But I'm not a practitioner, so educate me. How does LIDAR beat video?

jfoster 8y ago
I was quoting the article, but I think LIDAR is a bit more like having a 3D model of the world around the vehicle. The processing of a 1D array into a usable point cloud is not of concern at the frontier of autonomous vehicle research. My understanding is that the main reason Tesla would choose video instead is that cameras are super-cheap in comparison to LIDAR, and theoretically good enough if you have AI that can understand what the cameras are seeing. I have a feeling that Musk is pushing OpenAI toward researching the technology that Tesla will need to make their approach work. I hope they are successful; it is probably the fastest path toward ubiquitous autonomous vehicles. It's likely that Waymo can get theirs working faster, but they can't scale the distribution side as fast as Tesla could.