A predominant benefit of social living is the ability to share knowledge that cannot be gained through the information an individual accumulates based on its personal experience alone. Traditional computational models...A predominant benefit of social living is the ability to share knowledge that cannot be gained through the information an individual accumulates based on its personal experience alone. Traditional computational models have portrayed sharing knowledge through interactions among members of social groups via dyadic networks. Such models aim at understanding the percolation of information among individuals and groups to identify potential limitations to successful knowledge transfer. How- ever, because many real-world interactions are not solely pairwise, i.e., several group members may obtain information from one another simultaneously, it is necessary to understand more than dyadic communication and learning processes to capture their full complexity. We detail a modeling framework based on the simplicial set, a concept from algebraic topology, which allows elegant encapsulation of multi-agent interactions. Such a model system allows us to analyze how individual information within groups accumulates as the group＇s collective set of knowledge, which may be different than the simple union of individually contained information. Furthermore, the simplicial modeling approach we propose allows us to investigate how information accumulates via sub-group interactions, offering insight into complex aspects of multi-way communication systems. The fundamental change in modeling strategy we offer here allows us to move from portraying knowledge as a ＂token＂, passed from signaler to receiver, to portraying knowledge as a set of accumulating building blocks from which novel ideas can emerge. We provide an explanation of relevant mathematical concepts in a way that promotes accessibility to a general audience [Current Zoology 61 （1）： 114--127, 2015].展开更多
现有的网联自动驾驶车辆(Connected and Automated Vehicles,CAV)换道决策模型鲁棒性较差,存在安全隐患,且单纯依赖自车信息、较小范围内的感知信息,难以在CAV与人工驾驶车辆(Human-Driven Vehicles,HDV)混行的环境中推断出最优动作.综...现有的网联自动驾驶车辆(Connected and Automated Vehicles,CAV)换道决策模型鲁棒性较差,存在安全隐患,且单纯依赖自车信息、较小范围内的感知信息,难以在CAV与人工驾驶车辆(Human-Driven Vehicles,HDV)混行的环境中推断出最优动作.综合考虑感知信息、自车以及车-车通信(Vehicle-to-Vehicle,V2V)范围内上、下游CAV信息,提出一种混合交通流环境下集成多源信息融合的深度强化学习(Multi-Source Information Fusion Deep Reinforcement Learning,MSIF-DRL)端到端网联自动驾驶换道决策模型.首先,构建含有多源信息的状态空间,并为不同信息分配权重;其次,通过编码网络将各种动态信息编码到高维特征空间,进行信息融合得到特征图;然后,将其扁平化送入拥有优先经验回放机制的竞争双深度Q网络中,进行动作的选择和评估;最后,分别设计适用于主线、匝道CAV的奖励函数引导所提MSIF-DRL模型解决高速公路合流区驾驶场景中CAV的自由以及强制换道问题.基于SUMO软件在各种仿真条件下进行实验,将所提出的MSIF-DRL换道决策模型与现有换道模型进行比较,验证其有效性和优越性.研究结果表明:相较于现有模型,所提MSIF-DRL模型在各种仿真条件下均拥有最高的奖励值、换道成功率、合流成功率、平均行车速度、舒适性以及最低的碰撞风险,其中换道成功率、合流成功率、平均行车速度最大分别提升了29.17%、27.71%、17.43%;随着渗透率的提高,该模型在处理混合交通流环境下CAV的换道决策问题时拥有更强的性能和鲁棒性.展开更多
文摘A predominant benefit of social living is the ability to share knowledge that cannot be gained through the information an individual accumulates based on its personal experience alone. Traditional computational models have portrayed sharing knowledge through interactions among members of social groups via dyadic networks. Such models aim at understanding the percolation of information among individuals and groups to identify potential limitations to successful knowledge transfer. How- ever, because many real-world interactions are not solely pairwise, i.e., several group members may obtain information from one another simultaneously, it is necessary to understand more than dyadic communication and learning processes to capture their full complexity. We detail a modeling framework based on the simplicial set, a concept from algebraic topology, which allows elegant encapsulation of multi-agent interactions. Such a model system allows us to analyze how individual information within groups accumulates as the group＇s collective set of knowledge, which may be different than the simple union of individually contained information. Furthermore, the simplicial modeling approach we propose allows us to investigate how information accumulates via sub-group interactions, offering insight into complex aspects of multi-way communication systems. The fundamental change in modeling strategy we offer here allows us to move from portraying knowledge as a ＂token＂, passed from signaler to receiver, to portraying knowledge as a set of accumulating building blocks from which novel ideas can emerge. We provide an explanation of relevant mathematical concepts in a way that promotes accessibility to a general audience [Current Zoology 61 （1）： 114--127, 2015].
文摘现有的网联自动驾驶车辆(Connected and Automated Vehicles,CAV)换道决策模型鲁棒性较差,存在安全隐患,且单纯依赖自车信息、较小范围内的感知信息,难以在CAV与人工驾驶车辆(Human-Driven Vehicles,HDV)混行的环境中推断出最优动作.综合考虑感知信息、自车以及车-车通信(Vehicle-to-Vehicle,V2V)范围内上、下游CAV信息,提出一种混合交通流环境下集成多源信息融合的深度强化学习(Multi-Source Information Fusion Deep Reinforcement Learning,MSIF-DRL)端到端网联自动驾驶换道决策模型.首先,构建含有多源信息的状态空间,并为不同信息分配权重;其次,通过编码网络将各种动态信息编码到高维特征空间,进行信息融合得到特征图;然后,将其扁平化送入拥有优先经验回放机制的竞争双深度Q网络中,进行动作的选择和评估;最后,分别设计适用于主线、匝道CAV的奖励函数引导所提MSIF-DRL模型解决高速公路合流区驾驶场景中CAV的自由以及强制换道问题.基于SUMO软件在各种仿真条件下进行实验,将所提出的MSIF-DRL换道决策模型与现有换道模型进行比较,验证其有效性和优越性.研究结果表明:相较于现有模型,所提MSIF-DRL模型在各种仿真条件下均拥有最高的奖励值、换道成功率、合流成功率、平均行车速度、舒适性以及最低的碰撞风险,其中换道成功率、合流成功率、平均行车速度最大分别提升了29.17%、27.71%、17.43%;随着渗透率的提高,该模型在处理混合交通流环境下CAV的换道决策问题时拥有更强的性能和鲁棒性.
