Abstract: In order to improve the safety and comfort of heavy-duty truck intelligent driving decision-making systems, this paper decouples the horizontal and vertical decision-making, and proposes a behavioral decision-making method for hybrid trailers based on finite state machines and Markov decision processes. The horizontal decision-making uses a finite state machine, and the longitudinal decision-making uses a dual-mode Markov decision process with fixed probability and adaptive probability to achieve safe and efficient decision-making in park scenarios. The behavioral decision-making system adopts a hierarchical architecture, which is roughly short-term trajectory prediction-collision detection-horizontal and vertical behavioral decision-making-trajectory planning. This system is code implemented under the ROS1 framework, and is simulated and verified in the park scene in 51simone. The results show that the system can make reasonable and safe decisions in lane keeping, car following and lane changing scenarios.

Key words: tractor-trailer trucks; Markov decision process; behavior planning

摘要： 针对重型卡车因铰接结构导致运动复杂、横纵向耦合及决策难度大的问题，为提高重 型卡车智能驾驶决策系统安全及舒适性，文章对横纵向决策进行解耦，基于有限状态机与马 尔科夫决策过程提出混合拖挂牵引车的行为决策方法，其中横向决策采用有限状态机，纵向 决策采用固定概率和自适应概率的双模式马尔科夫决策过程，实现园区场景的安全高效决策。 行为决策系统采用分层架构的方式，大致为短期轨迹预测-碰撞检测-横纵向行为决策-轨迹规 划。本系统是在 ROS1 框架下完成代码实现，并在 51simone 中进行园区场景仿真验证，结果 表明该系统在车道保持、跟车及换道场景下均能做出合理及安全的决策。

关键词: 拖挂牵引车；马尔科夫决策过程；行为决策