Abstract: To handle the two shortcomings of existing methods, (i) nearly all models rely on high-definition (HD) maps, yet the map information is not always available in real traffic scenes and HD map ...