Data fusion between Image Segmentation and Google Maps Global Planner

In some cases, the data of Google Map may look confusing, in particular, the road path does not overlay on the satellite view. - - Moreover, the Bugcar does not have any means of high-resolution maps or high precision localization. The error of localizing the vehicle and the error of waypoint may lead to the vehicle running off-road!
Normally, human drivers use Google Maps for routing only; we do NOT use Google Maps to help us with local driving (keep the vehicle on the lane, avoid obstacles). In Bugcar, we do have the data of image segmentation that gives us the ego of the road with respect to the vehicle, so we may use the data to correct the localize and Google Maps' map error

As long as we keep the vehicle on the road, high precision localization is not important

So this idea, we assume that the car does well localization, and we try to interpolate/regress the waypoint depends on the vehicle's ego and the road's vector from image segmentation

tranqkhue / bugcar_archieved