It looks like it's not just using the camera/vnc images as input, but also uses depth data to simulate LIDAR input (http://deepdrive.io/ http://deepdrive.io/).

It looks like it's not just using the camera/vnc images as input, but also uses depth data to simulate LIDAR input (http://deepdrive.io/ http://deepdrive.io/). I wonder, what makes it hard to rely on solely the 2 cameras to calculate/guess the depth, similarly to the way us non-robotic drivers do it, when it comes to machine sensing. EDIT: found this: https://web.stanford.edu/class/ee368/Project_Autumn_1516/Reports/Appiah_Bandaru.pdf https://web.stanford.edu/class/ee368/Project_Autumn_1516/Rep... on the topic. Is there anything else I should read up on? EDIT: found lots of articles: https://www.google.com/webhp?sourceid=chrome-instant&ion=1&espv=2&ie=UTF-8#q=optical+flow+depth+estimation https://www.google.com/webhp?sourceid=chrome-instant&ion=1&e...