I thought the “needs lidar” debate was settled years ago? Lidar cannot read signs. It is also prohibitively expensive to put in vehicles. If you’re going to drive with a neural network you need as much training data as possible, which means as many sensors in as many vehicles as possible.

If your cameras detect something the lidar does not, you trust the cameras, every time. Lidar can very easily misinterperet the world. It works great for simple robots who need to know where walls are and don’t need to specifially identify animals, people, obstacles, speed bumps, construction zones, etc.

Theres also the simple fact that humans can drive just fine without having evolved a lidar sensor.