基于深度強(qiáng)化學(xué)習(xí)的智能網(wǎng)聯(lián)車輛盲區(qū)通行策略

打開文本圖片集

中圖分類號(hào)：TP273 文獻(xiàn)標(biāo)識(shí)碼：A DOI：10.3969/j.issn.1674-8484.2025.03.013

Abstract：Ablind spot pasing strategy method was proposed by using the deep reinforcement learning for intelligent connected vehicles （ICV）to prevent traficaccidents between vehiclesand pedestrians when passing through visual blindspots.Amathematical description model wasestablished for typical blind spot scenarios considering three indicators of safety，eficiencyand comfort; whileadeepreinforcement learning model was designedbasedonthe DoubleDQN（doubledeep Q-network）with theTTC （time tocolision）indicator to establishasetof physically interpretable reward functions，withtheoutputbeing the vehicle'sacceleratorand the brake pedal depth.Simulation experiments were conducted under three scenarios to assess thealgorithm eficacy.Theresultsshow thatthesimulationexperimentsverifytheeffctivenessof thealgorithm.Thecomfort is increased by more than 50% onaverage of this method，compared with the traditional DQN method.The method improvesdecision-makingaccuracy.Therefore，the longitudinaldecision-makingmethodachievesthe safety，theefficientand the comfortable.

Key words： inteligent connected vehicle （ICV）;deep reinforcement learning; pedestriancollisionavoidance; time to collision （TTC）

智能網(wǎng)聯(lián)汽車（intelligentconnectedvehicle，ICV）能夠增強(qiáng)行車安全，提高通行效率；因此，結(jié)合智能網(wǎng)聯(lián)技術(shù)的自動(dòng)決策和規(guī)劃方法在近年來(lái)成為了研究熱點(diǎn)[1-2]。（剩余10636字）