CPC G01C 21/3841 (2020.08) [G01C 21/1652 (2020.08); G01C 21/3811 (2020.08); G01C 21/3867 (2020.08); G01C 21/3878 (2020.08); G01C 21/3896 (2020.08); G06N 3/02 (2013.01)] | 31 Claims |
1. A method comprising:
computing, using one or more neural networks (NNs) and based at least on sensor data generated using one or more sensors of a vehicle, one or more outputs indicative of one or more locations in two-dimensional (2D) image space corresponding to one or more detected landmarks;
generating a distance function representation of the one or more detected landmarks based at least on the one or more locations;
generating a cost space by, for at least two poses of a plurality of poses of the vehicle represented in the cost space:
projecting one or more map landmarks corresponding to a map into the 2D image space to generate one or more projected map landmarks;
comparing the one or more projected map landmarks to the distance function representation;
computing a cost based at least on the comparing; and
updating a point of the cost space corresponding to an individual pose of the at least two poses based at least on the cost;
localizing the vehicle to the map based at least on the cost space; and
causing, at least in part and based at least on the localizing of the vehicle to the map, the vehicle to navigate along a trajectory.
|