Map road data generation model training method, related method, device and equipment

Abstract

The invention relates to a map road data generation model training method, a related method, a related device and equipment. The method comprises the following steps: respectively converting multiple groups of pre-training map element vector data obtained in advance into pre-training map element grating data; based on the multiple groups of pre-trained map element grating data and the corresponding map geometric grating data, pre-training a pre-constructed initial road generation model with increased randomness to obtain multiple over-fitting road generation models; according to the obtained multiple groups of map element grating data and based on the multiple over-fitting road generation models, multiple pieces of corresponding map geometric grating data are obtained; obtaining a training sample pair set; training a pre-constructed reward model for the set according to the training sample to obtain a trained reward model; and performing reinforcement learning training on the target road generation model group in parallel by using the trained reward model to obtain an optimal map road generation model.

Images