The invention discloses an autonomous decision-making method and system for an unmanned vehicle, a medium and a product, and relates to the technical field of unmanned vehicles, and the method comprises the steps: employing a supervised learning algorithm to train N environment model networks according to an offline data set, and obtaining N trained environment model networks; training a behavior strategy network by adopting an imitation learning algorithm according to the offline data set to obtain a trained behavior strategy network; training the target strategy network by adopting reinforcement learning to obtain a trained target strategy network: training the current target strategy network by adopting a training mode of interacting the current target strategy network and the trained environment model network in each round of training process of the target strategy network, environment model deviation and strategy deviation are introduced into rewards fed back by the environment; and adopting the trained target strategy network to make a decision for the operation of the unmanned vehicle. The reliability of unmanned vehicle decision making is improved.
本发明公开一种无人车自主决策方法、系统、介质及产品,涉及无人车技术领域,方法包括:根据离线数据集采用监督学习算法分别训练N个环境模型网络,得到N个训练好的环境模型网络;根据离线数据集采用模仿学习算法训练行为策略网络,得到训练好的行为策略网络;采用强化学习训练目标策略网络得到训练好的目标策略网络:在对目标策略网络的每轮训练过程中,采用当前目标策略网络和训练好的环境模型网络进行交互的训练方式训练当前目标策略网络,在环境反馈的奖励中引入环境模型偏差和策略偏差;采用训练好的目标策略网络为无人车运行进行决策。本发明提高了无人车决策的可靠性。
Unmanned vehicle autonomous decision-making method and system, medium and product
一种无人车自主决策方法、系统、介质及产品
2024-08-06
Patent
Electronic Resource
Chinese
IPC: | G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / B60W CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION , Gemeinsame Steuerung oder Regelung von Fahrzeug-Unteraggregaten verschiedenen Typs oder verschiedener Funktion |
European Patent Office | 2021
|Unmanned vehicle transverse dynamic avoidance decision-making method and unmanned vehicle system
European Patent Office | 2025
|