Deep Q learning for traffic simulation in autonomous driving at a highway junction
Deep Q learning for traffic simulation in autonomous driving at a highway junction
复制标题
DOI：
10.1109/smc.2017.8122738
复制
发表时间：
2017-10
期刊：
影响因子：
--
通讯作者：
K. Kashihara
中科院分区：
文献类型：
--
作者：
K. Kashihara
交通拥堵是一个严重的全球性问题。增强型或深度 Q 学习算法被应用于交通模拟研究。增强Q学习基于重复的局部搜索算法，能够在多智能体系统下找到最优路径。深度 Q 学习还能够考虑高速公路路口交通环境的动态变化，学习合适的策略。特别是Q学习的目标网络实现了损失函数的稳定调节。这种基于深度强化学习的智能方法可以成为优化汽车路径（包括自动驾驶系统）的有效工具。
Trafïîc congestion is a serious global problem. Enhanced or deep Q learning algorithms were applied to a traffic simulation study. The enhanced Q learning was based on a repeated local search algorithm, and it was able to find optimal pathways under a multi-agent system. Deep Q learning was also capable of learning a suitable strategy, considering dynamic changes in traffic circumstances at a highway junction. In particular, the target network of the Q learning realized a stable regulation of the loss function. This intelligent method based on deep reinforcement learning could become an effective tool to optimize car pathways including an autonomous driving system.