The invention discloses a lane keeping decision-making method based on deep reinforcement learning, which belongs to the field of automatic driving, does not need complex system modeling, and is simple in calculation method. The method can adapt to different road environments, and the output decision strategy can be adaptively adjusted according to state information and action information. The method mainly comprises the following steps: establishing a driving scene used by the lane keeping decision method; constructing a lane keeping model by utilizing a DDPG algorithm: interacting with an intelligent vehicle simulator TORCS environment, acquiring information data by a sensor, inputting the acquired information data as state space information, intelligently outputting action space information according to the state space information, obtaining a future state according to a network updating rule defined by the DDPG, and continuously iterating; setting a reward function, and calculating a return; designing a neural network architecture for lane keeping; designing algorithm hyper-parameters and noise exploration; and deep reinforcement learning is carried out through Markov decision. The method is mainly used for lane keeping of automatic driving.
本发明公开了一种基于深度强化学习的车道保持决策方法,它属于自动驾驶领域,其无需进行复杂的系统建模,计算方法简单;能够适应不同的道路环境,输出的决策策略能根据状态信息和动作信息进行自适应调整。它主要包括如下步骤:建立车道保持决策方法使用的行驶场景;利用DDPG算法构建车道保持模型:与智能车辆模拟器TORCS环境进行交互,传感器获取信息数据,所获取的信息数据作为状态空间信息输入,根据状态空间信息智能输出动作空间信息,再根据DDPG定义的网络更新规则得到未来状态,不断迭代;设置奖励函数,计算回报;设计车道保持的神经网络架构;设计算法超参数及噪声探索;通过马尔科夫决策深度强化学习。本发明主要用于自动驾驶的车道保持。
Lane keeping decision-making method based on deep reinforcement learning
基于深度强化学习的车道保持决策方法
12.12.2023
Patent
Elektronische Ressource
Chinesisch
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 / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen |
Automatic driving lane keeping decision-making method based on deep reinforcement learning
Europäisches Patentamt | 2025
|Hybrid automatic driving lane changing decision-making method based on deep reinforcement learning
Europäisches Patentamt | 2024
|DRNet: A Decision-Making Method for Autonomous Lane Changingwith Deep Reinforcement Learning
ArXiv | 2023
|