-
使用“嫦娥四号”任务实际场景进行实例分析和方法验证。
-
图8中,运用多选项作业对象来表示在停泊点A进行感知作业。其中,A点感知作业包含了感知行为和数据下传行为的组合,当对感知行为设置了“高码率试拍下传”分支点,则在数据下传行为中会相应的设置“高码率下传”分支点。这两个分支点与作业的同一选项绑定,一旦该选项被设置则两个分支点同时被任务规划系统自动选择。
-
图9中,建模的指令为一条真实指令:导航相机拍照,拍照目标角度和拍照方式均需计算确定。
其对应的指令序列流程为:〈试拍照,数据下传,正式拍照〉。其中,试拍照指令需要执行多少次可由对应的分支点确定。若此分支点的取值为3,则此指令序列将被自动展开包含3个试拍照指令的序列:〈试拍照,试拍照,试拍照,数据下传,正式拍照〉。
-
使用功能扩展后的PDDL对巡视器移动行为进行规划建模。首先在规划域描述结构中增加元标记“:processes”,对外部计算过程进行说明,并同时定义名如proc1和proc2的外部计算过程;在移动行为定义部分,运用这些外部计算过程来建模动作的动态性。图10中,使用(=?duration proc1)描述巡视器该行为的持续时长是由外部计算过程proc1负责每次迭代时重新计算;用(decrease energy proc2)描述巡视器剩余电量energy的降低数量由proc2计算。
-
假设巡视器当前位于S1点(坐标:–26.7,–1.2),科学家指定其移动到目标点(坐标:–45.1,–1.1)进行科学探测。当前点和目标点之间的其余导航点分别是S2点(坐标:–32.5,–0.9)和S3点(坐标:–37.9,–0.9),然后根据初始状态、目标状态以及测控约束、能源约束和光照约束等条件,启动任务规划计算,得到动作序列如表1所示。
表 1“玉兔二号”巡视器科学探测任务规划实例结果分析
Table 1.Analysis on the results of scientific exploration mission planning of Yutu 2 Rover
序号 动作 起始时间 结束时间 电能 备注 1 进测控区 12-23T22:31:50.0000 12-23T22:37:00.0000 100 S1(太阳方位角和高度角:–116.173 422,26.179 856) 2 感知 12-23T22:37:00.0000 12-24T01:50:00.0000 86 3 盲走移动 12-24T01:50:00.0000 12-24T02:10:00.0000 97 4 感知 12-24T02:10:00.0000 12-24T05:30:00.0000 76 S2 5 盲走移动 12-24T05:30:00.0000 12-24T05:50:00.0000 82 6 感知 12-24T05:50:00.0000 12-24T09:10:00.0000 100 S3 7 盲走移动 12-24T09:10:00.0000 12-24T09:30:00.0010 100 8 感知 12-24T09:30:00.0010 12-24T11:17:10.0010 100 S4(太阳方位角和高度角:–104.552 687,16.357 985) 9 出测控区 12-24T12:53:34.0010 12-24T12:57:44.0010 100 10 进测控区 12-24T23:41:50.0010 12-24T23:47:00.0010 100 11 探测 12-24T23:47:00.0010 12-25T03:22:50.0010 100
Research of a General Teleoperation Task Intelligent Planning Method
-
摘要:在地外天体执行遥操作任务时,在复杂约束条件下会出现多分支作业选择困难、事件属性设置复杂等现实难题。提出了一种通用型任务智能规划方法——分层规划对象模型(Hierarchical Plan Object Model,HPOM),巡视器在地外天体作业时,其分解为多选项作业、带约束行为、多分支指令序列、参数化虚拟指令4个层次,将带约束行为表示的计划转化为行为规划问题进行求解,获得求解方法集合。采用“人机协同迭代求解”(Human-In-The-Loop,HITL)的处理流程,生成指令序列以期实现对不同规划粒度方案的一致性验证。该方法已成功应用于“嫦娥四号”(Chang'E-4 ,CE-4)任务,为任务圆满成功提供了技术支撑。Abstract:Based on the technical challenges of teleoperation tasks in China, a general task intelligent planning method with hierarchical plan object model(HPOM)is proposed, which decomposes the task of rover into four levels: multi option operation, constrained behavior, multi branch instruction sequence and parameterized virtual instruction. The plan represented by constrained behavior is transformed into a behavior planning problem, and a set of solving methods is obtained to solve the practical problems such as difficult selection of multi branch operation and complex setting of event attributes under complex constraints. A process flow of human-in-the-loop(HITL)is proposed to verify the consistency of different planning granularity schemes and generate instruction sequence. This method has been successfully applied to Chang'E-4 mission, providing technical support for the success of the mission.Highlights
● The hierarchical plan object model(HPOM)is proposed. In the decomposition process,the concept of "branch point" is introduced,which affects the decomposition result and expands the flexibility of HTN planning. The software architecture is universal and is expected to be further applied in the subsequent complex tasks. ● In the hierarchical planning model,the method of transforming the planning problem with constraint behavior layer into PDDL planning problem is realized,which expands the idea of PDDL planning modeling. ● Based on HPOM,the human-in-the-loop(HITL)method under human-computer cooperation is explored,and an automatic method of mapping high-level objects to low-level objects is designed,which shortens the time for operators to build task plans,and realizes the complementary advantages of human experts and intelligent planning system. ● The method has been applied to chang'e-4 mission to support the success of the mission. -
表 1“玉兔二号”巡视器科学探测任务规划实例结果分析
Table 1Analysis on the results of scientific exploration mission planning of Yutu 2 Rover
序号 动作 起始时间 结束时间 电能 备注 1 进测控区 12-23T22:31:50.0000 12-23T22:37:00.0000 100 S1(太阳方位角和高度角:–116.173 422,26.179 856) 2 感知 12-23T22:37:00.0000 12-24T01:50:00.0000 86 3 盲走移动 12-24T01:50:00.0000 12-24T02:10:00.0000 97 4 感知 12-24T02:10:00.0000 12-24T05:30:00.0000 76 S2 5 盲走移动 12-24T05:30:00.0000 12-24T05:50:00.0000 82 6 感知 12-24T05:50:00.0000 12-24T09:10:00.0000 100 S3 7 盲走移动 12-24T09:10:00.0000 12-24T09:30:00.0010 100 8 感知 12-24T09:30:00.0010 12-24T11:17:10.0010 100 S4(太阳方位角和高度角:–104.552 687,16.357 985) 9 出测控区 12-24T12:53:34.0010 12-24T12:57:44.0010 100 10 进测控区 12-24T23:41:50.0010 12-24T23:47:00.0010 100 11 探测 12-24T23:47:00.0010 12-25T03:22:50.0010 100 -
[1] 吴伟仁,王琼,唐玉华,等. “嫦娥4号”月球背面软着陆任务设计[J]. 深空探测学报(中英文),2017,4(2):111-117.WU W R,WANG Q,TANG Y H,et al. Design of Chang'E-4 Lunar farside soft-landing mission[J]. Journal of Deep Space Exploration,2017,4(2):111-117. [2] 贾瑛卓,邹永廖,薛长斌,等. 嫦娥四号任务科学目标和有效载荷配置[J]. 空间科学学报,2018,38(1):118-130.doi:10.11728/cjss2018.01.118JIA Y Z,ZOU Y L,XUE C B,et al. Scientific objectives and payloads of Chang'E-4 mission[J]. Chinese Journal of Space Science,2018,38(1):118-130.doi:10.11728/cjss2018.01.118 [3] HOFFMANN J. Everything you always wanted to know about planning. KI 2011: advances in artificial intelligence[C]//34th Annual German Conference on AI. Berlin Heidelberg: Springer-Verlag, 2011. [4] 高薇,蔡敦波,周建平,等. 嫦娥三号“玉兔号”巡视器行为规划方法[J]. 北京航空航天大学学报,2017,43(2):277-284.GAO W,CAI D B,ZHOU J P,et al. Activity planning method for Chang' E-3 "Jade Rabbit" rover[J]. Journal of Beijing University of Aeronautics and Astronautics,2017,43(2):277-284. [5] DAUTENHAHN K. The art of designing socially intelligent agents:science,fiction,and the human in the loop[J]. Applied Artificial Intelligence,1998,12(7-8):573-617.doi:10.1080/088395198117550 [6] 贾阳,张建利,李群智,等. 嫦娥三号巡视器遥操作系统设计与实现[J]. 中国科学:技术科学,2014,449(5):470-482.JIA Y,ZHANG J L,LI Q Z,et al. Design and realization for teleoperation system of the Chang'E-3 rover[J]. Scientia Sinica Technologica,2014,449(5):470-482. [7] AGHEVLI A, BACHMANN A, BRESINA J, et al. Planning applications for three Mars missions with Ensemble[C]//The 5th International Workshop on Planning and Scheduling for Space. Baltimore: [s.n.], 2006. [8] CHIEN S. A generalized timeline representation, services, and interface for automating space mission operationss[C]//SpaceOps Conferences 2012. Stockholm: [s.n.], 2012. [9] BARREIRO J, BOYCE M, DO M, et al. EUROPA: a platform for AI planning, scheduling, constraint programming, and optimization[C]//The 22nd International Conference on Automated Planning and Scheduling(ICAPS). Sao Paulo Brazil: [s.n.], 2012. [10] 高薇,蔡敦波. 面向月面遥操作任务规划系统的搜索剪枝策略研究[J]. 航天控制,2017(4):73-78.GAO W,CAI D B. Search pruning strategy for mission planning in lunar teleoperation[J]. Aerospace Control,2017(4):73-78. [11] 叶培建,孙泽洲,张熇,等. 嫦娥四号探测器系统任务设计[J]. 中国科学:技术科学,2019,49(2):124-137.YE P J,SUN Z Z,ZHANG H,et al. Mission design of Chang'e-4 probe system[J]. Scientia Sinica Technologica,2019,49(2):124-137.