Short path planner training method based on natural confrontation environment generation

The invention relates to a short path planner training method based on natural confrontation environment generation, and the method comprises the following steps: collecting driving data, carrying out the grouping and analysis of the data, and obtaining the vehicle operation experience distribution;...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: JIAO XIWEI, XIE JUNPING, LIU PENG, ZHAO KEGANG, MAI HONGZAN
Format: Patent
Sprache:chi ; eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The invention relates to a short path planner training method based on natural confrontation environment generation, and the method comprises the following steps: collecting driving data, carrying out the grouping and analysis of the data, and obtaining the vehicle operation experience distribution; oTMA is identified through an empirical distribution utilization model, and manipulation of the OTMA is distorted, so that an antagonistic environment is constructed; and establishing a planner neural network by using a reward and punishment mechanism and putting the planner neural network into an adversarial environment for training. The environment arrangement provided by the invention is based on the natural behavior distribution of the vehicle, multiple dangerous operations of other vehicles are added in the distribution condition, and the correct vehicle is selected to perform the dangerous operation, so that the actual driving road condition is better displayed, and the path is more efficiently and accuratel