A framework for collaborative and adaptive MCM mission management - - PDF document

UDT 2020 A framework for collaborative and adaptive MCM mission management - Unmanned, Remotely Piloted & Autonomous Systems

A framework for collaborative and adaptive MCM mission management

A.L. van Velsen1, I. Mulders2, M.W.G. van Riet3

1Scientist, TNO, The Hague, The Netherlands, anna.vanvelsen@tno.nl 2Scientist, TNO, The Hague, The Netherlands, ingrid.mulders@tno.nl 3Scientist, TNO, The Hague, The Netherlands, martijn.vanriet@tno.nl

Abstract — This paper presents a mission management framework to support the planning, execution and evaluation

• f stand-off maritime mine counter measure (MCM) operations enabling collaborative behaviour. The mission

management framework, which is based on efficiency and effectiveness of the mission, is part of a multi-layer architecture that enables collaboration between unmanned vehicles. An increase in the efficiency and effectiveness of stand-off MCM operations is achieved on two different levels, task and mission level. This paper focuses on mission management, specifically for stand-off mine hunting, which distributes tasks over the available vehicles to optimize collaboration between the vehicles. Both a priori and adaptive mission planning of a mine hunting operation have been successfully demonstrated in open water.

1 Introduction

Many navies are moving to stand-off MCM for both mine hunting and mine sweeping. For mine hunting multiple heterogeneous unmanned vehicles are used (i) to carry out the separate tasks in the mine hunting process and (ii) to exploit the opportunity of parallel tasking to increase the efficiency of mine hunting operations. In contrast, in the legacy approach of mine hunting all tasks from detect to engage are conducted in a sequential manner. For the legacy approach planning, execution and evaluation are available but need to be modified to support the use of (multiple) autonomous systems. To utilize the collaboration potential in a system-of-systems for stand-

• ff mine hunting and to increase the safety and minimize

the workload on personnel, new planning and evaluation approaches are essential. In addition, the unmanned vehicles need to have a certain amount of autonomy to keep the stand-off mine hunting process effective and robust to, for example, the challenging communication conditions in underwater environments. To realize this level of autonomy van Vossen et al. (2020) [1] demonstrate various concepts of autonomous task execution while considering command and control, sensor management and navigation. There are multiple approaches to address the need for a mission management system. Current commercial off- the-shelf (COTS) mission management systems use a best- next-step approach to plan mine hunting tasks. However, using full mission awareness throughout the mission management process has the potential of improving the efficiency and effectiveness of the mission. For example, selecting a vehicle for a current task (such as detection, localization and classification (DLC)) that will not be needed for a task later on (such as identification). Therefore, the contribution of this paper is a novel mission management framework for collaborative autonomous vehicles for stand-off mine hunting operations. The framework can generate an a priori plan for the mission, which contains what needs to be done (tasks), by whom (resources), when (scheduling), where (area allocation), how good (performance metrics) and interdependencies between tasks.

2 The framework

A mission management framework is needed as mine hunting missions are challenging for several reasons. The mine hunting tasks depend on each other, which requires that the tasks are locally executed in a sequential order and that the unmanned vehicles exchange information. Additionally, several vehicles can work in parallel on the same task, for instance DLC, in different areas. Finally, the situational awareness in the mission area is limited beforehand and new information becomes available during mission execution. Therefore the mission management framework also allows for adaptive planning, in which the mission plan can be updated during mission execution, for instance when new information about the environment, the contacts, the vehicles or the mission itself (such as task progress) becomes available. 2.1 General description The mission management framework handles planning, execution and evaluation of the mission as shown in figure 1 and can optionally include adaptivity as well. A more detailed explanation of the three distinct blocks, which together form the mission manager, will be given in the next sections. The mission manager requires high level information such as the mission area, vehicle capabilities and estimated mine-like contact (MILCO) density to start the planning process. The planning block generates a mission plan and passes it on to the execution block, in which the mission plan is distributed to the vehicles. During mission execution task progress is reported back to the evaluation block in the framework in which the mission status and possible need for replanning are assessed taking into account both mission efficiency and effectiveness.

UDT 2020 A framework for collaborative and adaptive MCM mission management - Unmanned, Remotely Piloted & Autonomous Systems

Mission Management Framework

vehicle vehicle vehicle Execution Generate & Distribute task schedules Evaluation Performance & progress assessment Planning Area segmentation Update knowledge base Allocation of tasks and segments to vehicles task schedules task progress KB mission information

• Fig. 1. Mission management framework showing the planning,

execution and evaluation blocks. Mission information is needed to start planning. The generated plan is dispatched to the vehicles via task schedules in the execution block. And evaluation takes place before and during execution of the mine hunting operation. All relevant information is stored in an

• verall knowledge base (KB). Adaptivity is incorporated by

using the evaluation information to trigger replanning as indicated by the arrow connecting evaluation and planning.

2.2 Planning and execution The tasks on mission level are general tasks, which are divided into more detailed actions on vehicle level. The tasks that are planned on mission level are:  detection, localization and classification (DLC),  identification (ID),  transit,  safe depart or return and,  synchronization (forced information exchange) in case of adaptive planning. The planning of the mission is done using an hybrid method by combining two approaches, an expert knowledge based method linked to a goal-satisfaction

• planner. A goal-satisfaction planner will schedule actions

to try to reach the goal that is defined in the planning

• problem. By using the advantages of two different

approaches spatial and temporal planning can be combined. Three steps can be distinguished in the planning process as displayed in figure 1. The planner starts with segmenting the mission area in rectangular DLC and ID segments using a rule-based segmentation algorithm based

• n expert knowledge. The amount of segments is based on

the number of vehicles and their capabilities. Secondly, the central knowledge base (KB) of the mission management framework is updated with the available mission information and newly generated segments. The KB stores among others information about tasks, resources, situational awareness and courses of action. The use of a KB in the framework is the foundation for using full mission awareness during mission management. The KB is accessed during planning and during evaluation of the

• mission. During planning it serves as input for the next step

in the process: the allocation of tasks and segments to vehicles taking into account the dependencies between

• tasks. This part of the planning problem is described in a

STRIPS format (Stanford Research Institute Problem Solver) that is widely used for planning problems [2]. A mission plan is then created using a combinatorial problem solver that can deal with both logic and numeric variables [3]. A goal-satisfaction solver is selected in which tasks can be modelled with variables, preconditions, effects and goal-scores meaning that goal metrics can be taken into account. A goal metric is used to evaluate a plan quantitatively, not just on the goal state, by aiming for a plan with optimal performance with respect to the given metric such as mission duration or remaining risk. Mission planning in which mission duration is used as goal metric has been successfully demonstrated in an experiment on

• pen water mimicking a mine hunting operation as will be

discussed in section 3. Note that there is currently no method to assess the performance of the planner itself and to analyze whether a generated plan is optimal or not with respect to the given goal metric. Advantages of goal- satisfaction planners are the ease to scale the planning problem and the possibility to incorporate relevant information in the planning such as resource capabilities, environment parameters and performance models. During the first round of planning highly uncertain a-priori information is used, which can be updated during mission execution. In the next step, mission execution, the mission plan is distributed to the vehicles by generating a task schedule for each vehicle in which tasks are described in a specific

• rder. Task execution depends on the task’s position in the

task schedule and can depend on the availability of

• information. For example, an ID task can only start when

the vehicle has received MILCO information about the relevant area. 2.3 Evaluation Evaluation of the mine hunting operation is an important aspect of the mission management framework as most of the mission information has a high uncertainty. Environmental factors such as currents and seabed complexity are an example of highly uncertain mission information and can only be estimated using simple

• models. Another example is the number and location of the

MILCOs which is unknown and can only come available during execution of the mission. A mission plan is based

• n the currently available information and therefore the

quality of the plan depends on the quality of the

• information. To increase the quality of the operation

evaluation must be an ongoing process. During mission evaluation the mission information and mission progress are updated and the need for replanning is continuously

• assessed. Therefore, the framework contains an evaluation

method to estimate performance a-priori and during the

UDT 2020 A framework for collaborative and adaptive MCM mission management - Unmanned, Remotely Piloted & Autonomous Systems mission to monitor the efficiency and effectiveness of the mission. 2.4 Adaptivity Adaptive mission planning provides the ability to use in- situ information and to replan the remaining mine hunting tasks using this information. This will result in a higher efficiency and/or effectiveness. A mission plan is determined based on models of the environment, predicted vehicle performance and estimated MILCO distribution. These models might not suffice anymore when they deviate from reality and therefore the plan may need to be

• adapted. The evaluation functionality will provide

information on the necessity of creating a new plan or modifying the current plan. Therefore adaptivity is an essential component of the mine hunting mission management framework. In general there can be three reasons for replanning:

• 1. The actual performance or duration of a task

deviates from the planned performance or duration (comparison using measures

• f

performance (MOPs)). For example, the performance of the sonar is worse due to the presence of complex seabed structures or the identification task takes longer because the clarity

• f the water is poor.
• 2. New information is available such as the number

and location of the MILCOs after the DLC tasks have finished, which is communicated to all vehicles and to the mission manager during a synchronization task.

• 3. An unexpected event happens such as the

breakdown of a vehicle. In any of these cases a trigger is set which determines if the adaptive planning cycle is started. In this cycle the known current state is updated in the knowledge base and

• evaluated. If the old plan is infeasible or the performance
• f the new plan is significantly higher, the new plan will

be dispatched to all vehicles. This is not straightforward due to possible communication constraints.

3 Results

In October 2019 a trial was executed in open waters at lake Grevelingen, The Netherlands to demonstrate the collaborative autonomy using two heterogeneous lightweight autonomous underwater vehicles (AUVs) shown in figure 2. The first system is equipped for both DLC and ID, while the second vehicle can only perform

• ID. The goal of the scenario is to localize and identify all

four exercise mines in the area. In the experiment an area

• f 200 by 238 meter is assigned as mission area to the two
• vehicles. The mission information is sent to the planning
• functionality. In figure 3 a Gantt-chart and task maps of

the mission plan are shown. The Gantt-chart shows how the tasks are allocated to each vehicle in time, while the task maps, one for DLC and one for ID, illustrate how the vehicles are spatially planned. The autonomously generated plan has correctly assigned the DLC task to vehicle 1 for the whole mission area. After the DLC task, vehicle 1 is planned to carry out ID in the left half of the mission area (see figure 3), while vehicle 2 is planned to remain idle until after vehicle 1 has completed DLC, to then carry out ID in the rightmost half of the mission area. A synchronization task is planned after the DLC phase.

• Fig. 2. Two lightweight autonomous underwater vehicles

(AUVs).

During synchronization the mission manager receives information on the actual number of MILCOs that were found during DLC and that have to be identified. The number of MILCOs was deliberately underestimated in this experiment to demonstrate that a replanning phase would be triggered automatically. Due to the inaccurate mission information, describing a too low number of MILCOs in the area, replanning can improve the mission efficiency compared to the a priori plan. Consequently, replanning is triggered and a new plan is dispatched to the AUVs. The information exchanged during the synchronization task resulted in a list of MILCOs giving the location and classification. The new information on the location, initially the MILCOs were assumed to be uniformly distributed over the area, and an increase in the number of MILCOs resulted in the need for an update of the area segmentation and task allocation. Both the Gantt- chart and task maps of the new mission plan are shown in figure 4. The newly generated plan now contains four segments, which contain clusters of one or more MILCOs that need to be identified. Vehicle 1 is correctly tasked to first identify one segment after which it transits to the adjacent segment for another ID task. In the task map it is shown that vehicle 1 is tasked for identification in the leftmost two areas. In parallel, vehicle 2 is tasked to transit to the rightmost area where it will conduct ID, then transits to the adjacent area to perform its second ID task. Since all DLC tasks are finished, the task map for DLC is empty. The Gantt-chart shows that both vehicles have a longer ID task and a shorter one. This difference in estimated duration can be explained by the number of MILCOs that are to be identified in each segment (1 or 2). More MILCOs result in a longer estimated ID duration. In this example, the efficiency did not increase (total mission duration) but the prediction of the total mission duration is more accurate due to the use of newly obtained information in replanning. The example shows a proof of concept for adaptive planning using the mission management framework. It should be noted that the current scenario is quite simple with two vehicles in a small area with an initial assumption of uniform mine

• distribution. It is expected that adaptivity will become

more relevant and will result in increased performance and efficiency in a more complex scenario with a more challenging environment and a more realistic mine field.

UDT 2020 A framework for collaborative and adaptive MCM mission management - Unmanned, Remotely Piloted & Autonomous Systems

• Fig. 3. Mission plan for the experiment: Gantt-chart (top figure)

and task map for DLC showing one segment (bottom left) and task map for ID showing two segments (bottom right). The colour indicates which vehicle is performing the task: auv-1 in red, auv- 2 in cyan, standalone-master-1 in dark blue.

• Fig. 4. Mission plan after replanning: Gantt-chart (top figure) and

task map for DLC showing one segment (bottom left) and task map for ID showing four segments (bottom right). The colour indicates which vehicle is performing the task: auv-1 in red, auv- 2 in cyan. Note that the task map for DLC is empty as DLC has already been fully executed.

4 Conclusions

A mission management framework for collaborative autonomous stand-off mine hunting is implemented and

• demonstrated. The mission management framework

enhances collaboration between the vehicles and deals with planning, execution and evaluation. Spatial and temporal planning of tasks and vehicles is done using a hybrid method, which combines expert knowledge with a goal-satisfaction planner. The mission manager improves the performance of the mine hunting operation in several ways. Parallel task allocation can be used to decrease the total mission duration, while taking into account the required task order. Subsequently, plan performance is optimized by using a goal metric such as remaining risk in the goal-satisfaction

• planner. The mission management framework allows for

adaptivity: the efficiency and effectiveness of the mission plan are evaluated a priori and during mission execution, which can trigger replanning. In case of replanning updated information that is collected during mission execution is used to update the plan. A proof of concept of adaptive mission planning is demonstrated in a trial at lake Grevelingen, the Netherlands. An advantage of the developed mission management framework is the possibility to easily scale to more complex mission management problems by adding other tasks and system types. Moreover, the amount of environmental and progress information is limited in the current implementation, but can be extended to improve the performance prediction of the mission. Future work regarding the mission management framework involves exploring performance metrics and constraints to a larger extent such as planning with respect to remaining risk as described by van Vossen et al. [1] or taking into account mission duration constraints. Additionally, further work involves the challenges regarding timing of the evaluation and replanning due to the communication limitations that exist underwater.

References

[1] R. van Vossen, A.L.D. Beckers, J.J.M. van de Sande, Collaborative autonomy for naval stand-off mine countermeasure operations, UDT 2020, Rotterdam [2] D.S. Weld, AI Magazine 20, 2 (1999) [3] S. Russell, P. Norvig, AI: A Modern Approach, Ch10, Pearson, 3rd ed. (2010)

Author/Speaker Biographies

Ingrid Mulders received the MSc degree in Geophysics at the Utrecht University, The Netherlands. She is a Junior Research Scientist with the Acoustics and Sonar Department at the Netherlands Organisation for Applied Scientific Research (TNO), The Hague, The Netherlands, where she works on autonomous mine hunting and acoustic communications. Anna van Velsen received the MSc degree in Mechanical Engineering at the Eindhoven University of Technology, The Netherlands. She is a Research Scientist with the Acoustics and Sonar Department at the Netherlands Organisation for Applied Scientific Research (TNO), The Hague, The Netherlands, where she works on autonomous mine hunting. Martijn van Riet received the MSc degree in Astrophysics at Leiden University, The Netherlands. He is a Senior System Engineer with the Acoustics and Sonar Department at the Netherlands Organization for Applied Scientific Research (TNO), The Hague, The

• Netherlands. He has developed systems for ASW and

MCM and is now working on autonomous mine hunting.