基于蒙特卡洛搜索树的自动飞行机动
详细信息 查看官网全文
- 英文论文题名：Monte-Carlo Tree Search based Aerial Maneuvering
- 论文作者：陈向 ; 王维嘉 ; 魏文领 ; 朱雪耀
- 英文论文作者：Xiang CHEN ; Weijia WANG ; Wenling WEI ; Xueyao ZHU ; Flight Automatic Control Research Institute
- 年：2016
- 作者机构：中航工业西安飞行自动控制研究所;
- 论文关键词：飞行机动 ; 飞行控制 ; 多目标优化 ; 蒙特卡洛搜索树
- 英文论文关键词：aerial maneuver ; flight control ; multi-objective optimization ; Monte-Carlo Tree Search
- 会议召开时间：2016-11-25
- 会议录名称：2016年航空科学与技术全国博士生学术论坛摘要集
- 语种：中文
- 分类号：E926.3
- 学会代码：XBHH
- 会议名称：2016年航空科学与技术全国博士生学术论坛
- 会议地点：中国陕西西安
- 主办单位：西北工业大学研究生院、中国高校航空学院院长联席会
- 学会名称：西北工业大学航空学院
- 页数：1
- 文件大小：565k
- 原文格式：D
- 会议级别：全国
摘要
本文研究了无人机空战背景下的自动飞行机动问题。作者将自动飞行机动问题抽象为一个马尔可夫决策过程(MDP),并提出了在多目标蒙特卡洛搜索树(MOMCTS)框架下解决该问题的构想。基于某型战斗机仿真模型,本文以筋斗机动为例对所提出方法进行了实验验证,并将实验结果与传统方法(查表式参考轨迹生成+PID控制律轨迹跟随)进行了对比。实验结果表明,在消耗更多的计算资源的代价下,基于多目标蒙特卡洛搜索树的控制器生成的机动轨迹较传统方法在前飞距离和机动始末状态高度差等指标中的表现均更加优越,并获得了接近于职业战斗机飞行员能力极限的机动飞行轨迹。
This paper studies the problem of loop maneuvering control for future UAV air combat application. The authors formulate the problem as a Markov Decision Process(MDP), which is solved in the Multi-Objective Monte-Carlo Tree Search(MOMCTS) framework, with three to-beoptimized objectives loop Success, span and dH and the action space is defined by discrete elevator angles. The proposed method is validated by simulation based on the aerodynamic model of a typical fighter aircraft. The performance of MOMCTS approach is compared with traditional trajectory generation and PID control law approach. Results show that MOMCTS acquires a close-to-human maneuver trajectory, and outperforms the baseline method in both span and dH indicator, at the cost of requiring more computational resources.
This paper studies the problem of loop maneuvering control for future UAV air combat application. The authors formulate the problem as a Markov Decision Process(MDP), which is solved in the Multi-Objective Monte-Carlo Tree Search(MOMCTS) framework, with three to-beoptimized objectives loop Success, span and dH and the action space is defined by discrete elevator angles. The proposed method is validated by simulation based on the aerodynamic model of a typical fighter aircraft. The performance of MOMCTS approach is compared with traditional trajectory generation and PID control law approach. Results show that MOMCTS acquires a close-to-human maneuver trajectory, and outperforms the baseline method in both span and dH indicator, at the cost of requiring more computational resources.
引文