The invention discloses a multi-modal scene driving behavior modeling method and system based on inverse reinforcement learning, relates to the technical field of intelligent driving, classifies scenes of training data based on model parameters through a self-supervised contrast learning algorithm, and outputs a scene label corresponding to each training data. And taking the data of the same scene label as a group, and circulating the inverse reinforcement learning-scene clustering process again until a stable model parameter-training data pair is obtained. And training a global cost function by using'model parameter-training data 'pairing, so that the global cost function can judge a scene corresponding to the vehicle according to the state of the vehicle and output a corresponding model parameter. The problem that a current driving behavior modeling method for applying inverse reinforcement learning to intelligent driving cannot adapt to a complex mixed interaction scene due to limitation of a linear reward function is solved.
本发明公开了基于逆强化学习的多模态场景驾驶行为建模方法及系统,涉及智能驾驶技术领域,通过自监督对比学习算法基于模型参数对训练数据的场景进行分类,输出每个训练数据对应的场景标签。以相同场景标签的数据为一组,再次循环“逆强化学习‑场景聚类”过程,直到得到稳定的“模型参数‑训练数据”配对。使用“模型参数‑训练数据”配对训练全局代价函数,使其能根据车辆的状态判断其对应的场景并输出对应的模型参数。解决了目前将逆强化学习应用于智能驾驶中的驾驶行为建模方法由于受到线性奖励函数的限制,无法适应复杂的混编交互场景的问题。
Multi-modal scene driving behavior modeling method and system based on inverse reinforcement learning
基于逆强化学习的多模态场景驾驶行为建模方法及系统
2024-08-30
Patent
Electronic Resource
Chinese
IPC: | 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 / G06F ELECTRIC DIGITAL DATA PROCESSING , Elektrische digitale Datenverarbeitung / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen |
Automatic driving key scene generation method based on multi-agent reinforcement learning
European Patent Office | 2024
|Automatic driving solution under multi-target complex traffic scene based on reinforcement learning
European Patent Office | 2022
|