Fundamental questions raised about practicalities of 3D mapping.
Each driverless car company is busy creating gargantuan quantities of 3D maps so that their driverless cars can navigate safely. The better the library of pre-existing 3D maps, the less the driverless car has to think on the spot, so to speak. In other words, the driverless car simply has to process any differences there might be between the map library and the reality on the road. This is a much smaller task for the individual car. But the sheer amount of data needed means that it is difficult and costly to collect, collate and maintain. Moreover, it is grossly inefficient for each car company to produce its own mapping. As yet, there is little sign of collaboration between the companies. There is a bigger issue too. Humans get around perfectly well with simple 2D maps. Nature has designed them that way. A truly intelligent robot driver would do the same thing. This would do away with the need for complex 3D mapping altogether.
Link to article: https://www.ft.com/content/2a8941a4-1625-11e8-9e9c-25c814761640
You may also like to browse other AI posts: https://www.thesentientrobot.com/category/ai/