Available online at www.sciencedirect.com Available online at www.sciencedirect.com
ScienceDirect ScienceDirect
Procedia Computer Science 00 (2017) 000–000 Available online at www.sciencedirect.com Procedia Computer Science 00 (2017) 000–000
ScienceDirect
www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia
Procedia Computer Science 112 (2017) 534–543
International Conference on Knowledge Based and Intelligent Information and Engineering SysInternational Conference Knowledge Based and Intelligent Information tems, on KES2017, 6-8 September 2017, Marseille, Franceand Engineering Systems, KES2017, 6-8 September 2017, Marseille, France
A task-oriented and parameterized (semi) autonomous navigation A task-oriented and parameterized (semi) autonomous navigation framework for the development of simulation systems framework for the development of simulation systems Juliana R. Brondaniab *, Edison P. de Freitasab , Luis A. L. Silvabb ab ab Juliana R. Brondani *, Edison P. de Freitas , Luis A. L. Silva
Graduate Program in Electrical Engineering, Federal University of Rio Grande do Sul, Av. Paulo Gama, 110, Porto Alegre, 90040-060, Brazil. b Graduate Program in Electrical Engineering, Federal University ofof Rio Grande do Sul, Paulo 1000, Gama,Santa 110, Porto 90040-060, Graduate Program in Computer Science, Federal University Santa Maria,, Av. Av. Roraima, Maria,Alegre, 97105-900, Brazil.Brazil. b Graduate Program in Computer Science, Federal University of Santa Maria,, Av. Roraima, 1000, Santa Maria, 97105-900, Brazil.
a a
Abstract Abstract Agent behaviors in simulation systems are related to fundamental capabilities of realistically developing (semi) Agent behaviors in simulation systems are related to fundamental of realistically developing (semi) autonomous navigation actions. This is particularly important when dealing capabilities with the implementation of Computer Generated autonomous navigation actions. This isin particularly important when dealing with the implementation Computer Generated Forces (CGFs) for simulation systems tactical military training applications. Moreover, these systemsoftake into consideration Forces (CGFs) for of simulation systems in tactical military Moreover, these systems takeoninto consideration the particularities the domain-specific simulation taskstraining and the applications. numerous heterogeneous CGFs inserted them in order to the particularities of the domain-specific and the numerous heterogeneous in order to generate better knowledge and learning simulation experience tasks to simulation system users. Based on CGFs these inserted reasons, on thisthem paper reviews generate knowledge and aslearning experience to simulation system users. (semi) Based autonomous on these reasons, this paper reviews recurrent better navigation problems to propose a task-oriented and parameterized navigation framework to recurrent problems as in to propose task-oriented and parameterized autonomoustechniques, navigationand framework to deal with navigation CGF navigation needs military asimulation. Combining global and (semi) local navigation controlled deal with between CGF navigation needs in military simulation. Combining global and navigation techniques,ofand controlled transition alternative degrees of navigation autonomy, the framework aimslocal to overcome the challenges implementing transition between degrees ofand, navigation autonomy, theallow framework aimswith to overcome theand challenges of implementing customizable CGF alternative navigation behaviors at the same time, to interaction both users other simulation systems customizable navigation behaviors same time, to allow with both users and otherinsimulation systems in distributed CGF simulation settings. A caseand, studyatisthe presented in which theinteraction proposed techniques are analyzed a domain-specific in distributed simulation settings. A case is presented whichthe thestudied proposed techniques are analyzed in a domain-specific simulation problem providing evidence of study their suitability to in address military simulation problems. simulation problem providing evidence of their suitability to address the studied military simulation problems. © 2017 The Authors. Published by Elsevier B.V. © 2017 2017 The The Authors. Authors. Published Published by by Elsevier Elsevier B.V. B.V. © Peer-review Peer-review under under responsibility responsibility of of KES KES International. International Peer-review under responsibility of KES International. Keywords: Navigation problems; task-oriented navigation; semi-autonomous algorithms; simulation systems Keywords: Navigation problems; task-oriented navigation; semi-autonomous algorithms; simulation systems
* Corresponding author. Tel.: +55 (55) 991016539. * Corresponding Tel.: +55 (55) 991016539. E-mail address:author.
[email protected] E-mail address:
[email protected] 1877-0509 © 2017 The Authors. Published by Elsevier B.V. 1877-0509 ©under 2017responsibility The Authors. of Published by Elsevier B.V. Peer-review KES International. Peer-review under responsibility of KES International.
1877-0509 © 2017 The Authors. Published by Elsevier B.V. Peer-review under responsibility of KES International 10.1016/j.procs.2017.08.161
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
535
1. Introduction In the field of Artificial Intelligence (AI) aiming to model realistic objects’ behaviors, simulation systems stand out as virtual environments for the investigation of solutions for navigation problems. Simulation systems relying on different navigation techniques are commonly used in professional education settings, where the most prominent scenario is the case of simulation-based training in the military1,2,3. The popularity of such systems is due to the fact that they allow users to recreate real-life problem-solving situations without endangering people and spending valuable (and often scarce) resources. In addition, operational analysts can rapidly specify and evaluate both existing and new systems and procedures in the underlying simulations. In the military, Computer Generated Forces (CGFs)4 are crucial components of such simulation systems since these forces can show key real-life navigation behaviors in simulation exercises. For this reason, CGFs have, as a primary goal, to replicate relevant aspects of either human or equipment behaviors, or both, while following key tactics and procedures of a military doctrine. As such simulation systems are often used as tools for training, they also offer the possibility of human interaction to some extent in the navigation algorithms, which can be reflected as semi-autonomous CGF implementations. To handle (semi)autonomous navigation behaviors in simulation applications, deliberative and reactive navigation algorithms5 from AI usually address isolated navigation issues, seldom considering the modeling and implementation requirements of a complex task-oriented simulation scenario. As investigated in this work, navigation algorithms for virtual tactical simulation systems need to combine global and local navigation techniques, which are designed to address layered simulations where static and dynamic navigation and collision detection/avoidance issues are present. Navigation algorithms also need to provide intelligent behaviors that are consistent with real life military commander decisions, even considering incomplete knowledge about the battle-situations being simulated. These behaviors can be directed to simulate groups of military units, which can be taken as either individuals or aggregates formed of heterogeneous pieces of military equipment. In effect, such algorithms ought to address navigation tasks in which simulated units are capable of executing domain-specific behaviors over large virtual terrain representations according to a given military doctrine. In this context, complementary navigation techniques are rarely presented in a combined manner in a task-oriented navigation framework, i.e. tackling different problems at once, so that realistic simulation scenarios can be handled. Observing this landscape, this paper presents a hybrid (semi) autonomous navigation framework for the implementation of task-oriented intelligent behaviours for CGFs used in military simulation systems. The framework combines both deliberative and reactive navigation algorithms, which are parameterized in different ways to consider the needs of heterogeneous military units being simulated and to produce realistic navigation solutions for them. In addition, the framework also considers the tactical and doctrinal needs of military simulations, like coordinated movement, aiming to become a parameterized (semi) autonomous solution to be reused in other applications with similar tactical navigation challenges. The framework is based on a real simulation scenario involving simulations of mobile artillery batteries in the context of a project aiming the design and prototype of a distributed virtual tactical simulation system for the Brazilian Army: the SIS-ASTROS project. This particular problem scenario is a realistic test-bed application for the design and evaluation of the framework since the virtual tactical simulation characteristics of such artillery battery units capture the most relevant (semi) autonomous navigation issues previously mentioned. The paper is organized as follows: First, prominent problems for (semi)autonomous agent navigation are discussed. Second, the framework to approach navigation issues in (semi)autonomous simulation systems for tactical military training is proposed. Third, a case study is presented in which the proposed implemented solutions are analyzed. Fourth, the paper proposals are contrasted with related works. Finally, a discussion regarding the contributions of the framework for the development of military simulations is provided along with a description of future works. 2. Navigation problems of simulation systems Simulation systems are broadly used in the military training scenario as to familiarize trainees with tools, vehicles, procedures and doctrines explored in complex military operations. Besides the individual usage of simulation approaches, which are traditionally classified as live, virtual and constructive6, there is an increasing interest on systems that combine these simulation types, leading to a “blended training”7 approach. In this context, the exploration of computational techniques to support the movement of agents from one point to another in a virtual terrain has been
536
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
discussed in the AI literature5. In a simulation system used for virtual tactical military training, however, realistic taskoriented navigation solutions ought to deal with different issues. First, CGFs should be able to follow a plan or a doctrine, adapting their expected behaviors to the execution of different simulation tasks. A military doctrine, for example, states a set of procedures that are believed to be the best way (sometimes the mandatory way) to approach military situations. In contrast with usual agent implementations in computer games, CGFs in simulation systems cannot deviate too much from what is established by such best practice domain rules. Otherwise, users would be exposed to wrong training experiences, leading to the formation of undesired military skills. Second, the degree of autonomy of CGFs may not be characterized as fully autonomous in blended constructive and virtual simulations. In educational simulation environments, agents may need to have the capacity of receiving both computational and human inputs, in a scenario that requires the use of either autonomous or semi-autonomous navigation algorithms, or both combined. Therefore, these kinds of simulation systems may involve CGFs with alternative levels of autonomy as defined by the SAE International Standard8. It means that autonomy degrees may range from the planning or execution of simulation actions, presenting characteristics that vary from completely manual operation in the simulations, passing by the execution which considers user interference, until the simulation operation which is completely automated. Although these semi-autonomy capabilities are a recurrent topic in the field of robotics, especially in the field of autonomous driving like in the General Motor’s EN-V city vehicle9, relevant differences between them and the simulation scenario indicate that there is a lot to be explored as far as semi-autonomy is concerned in simulation-based training. Third, virtual terrains represented in simulation systems should reflect real-life terrain characteristics as to support the development of military navigation exercises, where realistic terrain representations are likely to be rich in detail and large in size. As far as the size and complexity of the terrain and the quantity of CGFs moving on it are concerned, planning a doctrine-based navigation path while avoiding collisions with static and dynamic obstacles can quickly become prohibitive to be calculated in real time simulation executions. To provide realistic simulations, the navigation algorithms should also take into account the characteristics of each simulated unit, for instance, as not all of them may be able to execute the same types of movement tasks as to follow a given military doctrine. Finally, navigation behaviors directed to CGFs representing military aggregates are often related to formationoriented movements, which can be defined as coordinated movements. A variety of formation control characteristics are analyzed in10, where the analysis results are categorized into position, displacement, and distance-based control properties. As these categories are usually explored around applications turned to the field of robotics, they tend to be dependent on what it is possible to sense and process according to the underlying robotic hardware, while control strategies in simulation systems are much more focused on training resources provided by simulation systems to their users. To sum up, these are major problems to consider when designing of parameterized (semi) autonomous agent navigation algorithms to simulated environments, particularly considering military simulations. 3. The proposed task-oriented and parameterized (semi) autonomous navigation framework A schematic architecture representation involving distributed simulation systems11 is presented in Figure 1. There, multiple simulation systems can be running during a simulation exercise (Figure 1(A)). Then, overall simulations emerge from the individual simulations executed in each one of these interconnected systems. The simulation component of this architecture which is responsible for the overall navigation autonomy control capabilities (Figure 1(B)) communicates and receives local (dotted arrows) and global (solid arrow) simulation feedback from all involved simulation system components. According not only to such simulation status feedback, but also to simulation inputs from both users and other simulation systems, the level of simulation autonomy can be switched in different moments of the simulation exercises. In this architecture, each simulation exercise involves the execution of a sequence of domainspecific navigation tasks, which are hierarchically decomposed in subtasks (Figure 1(C)) in a task decomposition structure12, as to achieve a higher simulation objective. In the navigation simulations, these tasks control the heterogeneous units involved (Figure 1(D)), returning a simulation feedback for the execution of movement-related actions as long as such doctrine-based simulation tasks are being executed. In convoy-types of formation inserted in the navigation simulations (Figure 1(D)), the existence of heterogeneous CGFs implies that each convoy unit follows its own state control machine (Figure 1(E)). In doing so, a certain domain-specific task relies on parameterized navigation
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
537
behaviors which are consistent with the type of simulated unit that is executing the task. Finally, these components act on a hierarchical representation of a virtual terrain (Figure 1(F)) as to capture the realistic details and dimensions of real-life geographical regions.
Figure 1 – Schematic representation of the architecture of the proposed framework
3.1. Task-oriented and parameterized components of the framework architecture Military aggregations are often formed of individual units having heterogeneous natures and functions (Figure 1 (D)). In other words, these units belonging to aggregated CGFs can have different physical and tactical characteristics having a crucial influence on their expected navigation behaviours. These characteristics dictate CGF behaviours like the choice of a given convoy formation organization used in certain domain-specific navigation tasks, the degree of autonomy the users want to experiment in the navigation simulations, etc. For these reasons, the proposed framework uses parameterized algorithms in which control parameters range from a fine-grain level of detail, as low as the dimensions and manoeuvre limitations of a military vehicle, to coarse-grain definitions, such as the kind of formation military users want to simulate. The framework also relies on a task-oriented solution due to domain-specific tactical needs of military simulations. In13, tactics are described as “a specification of responsive, goal-directed behaviours” as these behaviours “must be able to respond in a timely manner to events that interfere in the achievement of its goals”. This description fits the navigation problem considered here, as military doctrine and tactics regulate the simulation exercises. To deal with such complex CGF simulation behaviours, domain-specific simulation tasks are organized in hierarchies where a complex task is decomposed in a set of simpler ones (Figure 1 (C)). Besides allowing the reuse of generalized task-oriented implementations, this decomposition considers the following possibilities: different tasks being simultaneously executed; their asynchronous execution; conditional execution; and the treatment of exceptions. 3.2. Autonomy control components of the framework architecture Semi-autonomous systems are shaped by autonomous behaviors that require some degree of human intervention as to complete a task14. In contrast with fully autonomous systems, the combination of different levels of autonomy benefits the training of users as they can be inserted in the simulation loop. As presented in Figure 1 (A), both domain users and other simulation systems are allowed to participate in the decision-making process as simulation exercises
538
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
are executed. In the proposed framework, the presence of human intervention and assistance in the simulation exercises is implemented in different components of the overall simulation architecture. For example, during a phase of pathfinding planning, users can directly determine tactical areas in the virtual terrain that are preferred. They can also input simulation parameter values that may change the heuristic of the pathfinding algorithm according to the needs of the domain-specific simulation tasks being executed. Moreover, the switch of autonomy in the simulation system from either manual to semi-autonomous or to autonomous and vice versa should be carefully controlled in the simulations. Abruptly changing the control of a part of the navigation simulation task may cause errors, compromising the entire simulation realism. In addition, if users or other simulation systems inserted in the simulation execution loop fail to answer to a simulation request, the control components of the simulation systems need to identify this failure and supply an autonomous course of action for a CGF as to not compromise the rest of the simulation exercises. To do so as proposed in the framework, each possible mode of autonomy, distributed between manual, semi-autonomous and autonomous, is modeled as a set of states in finite state machine structures as part of the autonomy switch control component of the proposed framework architecture (Figure 1(B)). 3.3. Global and local navigation components of the framework architecture Autonomous global and local navigation for CGFs acting in virtual environments is a relevant issue for simulation systems, where techniques like A* algorithms, potential fields and their variations are explored in the literature5. For an autonomous and semi-autonomous environment shaped for domain-specific military simulations as aimed by the framework, however, these techniques have to be adapted to fit a more realistic simulation scenario. In effect, the large virtual terrain represented in the simulation environment portraits the reality. In them, CGFs may need to cross areas that are pre-processed and therefore described in a map representation in the simulation system. These units may also cross areas which have to be mapped at simulation run time. Consequently, three main topics are considered by the global and local navigation control framework components (Figure 1 (C) and (E)): navigation planning considering pre-processed terrain information, identified as global navigation; steering in virtual terrain areas that need to be processed during simulation execution, identified as local navigation; and representation of large virtual terrains (Figure 1(F)) so that the real time execution of these complementary navigation strategies be possible. To tackle the terrain size problem, the proposed framework explores a complete Quadtree data structure15 in the virtual terrain representation. In contrast with standard commercial terrain representations exploring flat representation solutions, every node in the tree in this Quadtree structure, with the exception of its leaf nodes, has four children nodes. Moreover, all leaves are on the same depth level. With this Quadtree representation, key terrain characteristics are divided in various levels of abstraction. Among other benefits, this hierarchical solution allows the CGF algorithms to quickly access information present in a certain node in a certain level, allowing to eliminate the search of a path in large chunks of the virtual terrain, optimizing then the overall pathfinding process. For planning the best route for CGFs to traverse this virtual terrain representation (Figure 1(F)), the A* algorithm is initially used as it provides optimality and completeness for global navigation. However, this type of algorithm has a high computation cost when simulation exercises are executed. For this reason, the planner algorithm is applied in the hierarchical Quadtree structure15. Thus, the pathfinding algorithm starts analyzing the coarser level of the terrain representation, determining which nodes are part of the pathfinding solution. After that, the algorithm executes again in a finer level terrain representation, now considering only the nodes that are children of those returned in the previous execution step. After the navigation path is planned, the CGFs navigate the resulting terrain nodes. When doing so, they also consider dynamic obstacles like other agents that are moving during the simulation executions. To deal with these dynamic navigation aspects, a local steering navigation approach is used in the CGFs implementations using a navmesh16 representing small terrain areas while simulations are executed. Then, the steering algorithm avoids collisions with other dynamic obstacles mapped in these local terrain representations. 3.4. Convoy coordinated movement components of the framework architecture A formation is a complex group of units, which has an orientation (a front, a back, etc.) and each unit in the formation maintains a relative position with respect to the other units in the group17. Moreover, complex behaviors emerging from such a group are observed when the heterogeneous group members (Figure 1 (D)) need to move at the
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
539
same speed, take the same path, and arrive at the same time at a given position. From this landscape, the proposed framework relies on a formation control approach that organizes CGF aggregates according to either a displacement from a center point in the formation or a commander unit, which are taken as parameters in the task-oriented navigation algorithms (Figure 1 (C)). When convoy coordinated movement is required, such a commander unit is elected so that the simulation time could be optimized. In doing so, the units’ positions in the virtual terrain are used to calculate a coarse-grained pathfinding for the whole CGF group. While the group is moving according to a resulting path, individual units in it still have to deal with speed control and group cohesion issues when dealing with static and dynamic obstacles. To tackle such problems, the framework relies on force-based concepts18, with which CGFs are kept in the formation due to cohesion rules, as for agents to stay close to each other; separation rules, as for agents to avoid collision with others; and alignment rules, as for agents match their velocity to neighbors. 4. A virtual tactical simulation case study for the proposed framework This case study is inserted in a research and prototyping project being developed at the Federal University of Santa Maria, Brazil – the SIS-ASTROS project. Taking the artillery battery as the object of the performed CGF navigation behavior implementations, the tactical simulations are focused on the choice of battery positions in a battle scenario, along with the organized battery occupation of these positions. In such problem, military actions aim to move the battery in between different positions in a battle theater. In this scenario, the framework can be illustrated by a battery task regarding the “occupation of a waiting position”. To complete it, the recommended actions are: i) starting from a given position in terrain, a selected set of artillery battery units move in a convoy formation to the waiting position. To do so, the location of this waiting position is selected by training users according to the battery missions to be executed; ii) on arrival at the waiting position, the military personnel belonging to the battery recognize the waiting position, finding the particular locations that each battery vehicle should park. When these individual locations are chosen, the battery units break formation and each vehicle move independently to its defined location. An example of the battery vehicles arriving at the waiting position is shown in Figure 2. In Figure 2 (A), the battery is initially moving through roads in a line formation, where these roads are pre-processed in the virtual terrain maps used by the simulation system. Figure 2 (A) also shows that the vehicles are leaving the road as to arrive at a waiting position selected by training users. After leaving the road as to reach the vehicles’ destinations, they need to go through a terrain area that was not previously mapped by the simulation system. Figure 2 (B) presents the situation in which the battery vehicles are arriving at the waiting position, where they are still aggregated while moving in a line formation to reach their destination. Figure 2 (C) shows the 2D representation of the waiting position arrival situation, where the aggregated vehicles have a single symbolic representation (a blue rectangle). Figure 2 (D) shows the tactical 2D map representation of the battery vehicles at the waiting position, where Figure 2 (E) shows them parked inside this position, as they have broken the convoy formation to move independently as to reach their particular park positions. This is also presented in the 2D representation of Figure 2 (D), where the battery is not represented as a military aggregate (there are several blue rectangles). In this virtual tactical simulation system, a fundamental simulation characteristic is that the 2D views as seen in Figure 2 (C) and (D) are closely connected to their corresponding 3D visualizations. In the 3D views in Figure 2 (A), (B) and (E), different battery vehicles are shown, supporting the need for parameterized algorithms to tackle these heterogeneous units in the simulation exercises. 4.1 A task-oriented and parameterized solution for virtual tactical simulations The proposed task-oriented and parameterized algorithms can be illustrated in this case study by using this waiting position occupation task. To complete this task, the battery convoy has to finish all the following subtasks: i) step 1: Using a selected military formation from a set of formations accessible via parameters in the task-oriented navigation algorithms, the control algorithm moves a selected artillery battery convoy to the closest road from a given initial position in the scenario of a battle situation; ii) step 2: Using a calculated path, which is based on the processing of road representations available in the virtual scenario (a graph involving these roads), the navigation algorithm moves the artillery battery convoy according to a selected military formation to reach a road location which is closer to the waiting position; iii) step 3: Once a position closer to the waiting position is reached, the convoy leaves the road. So, the navigation algorithm moves the artillery battery convoy according to a selected military formation to the border
540
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
of the waiting position (Figure 2 (A)); iv) step 4: Breaking the military formation, the navigation algorithm independently moves each artillery battery vehicle according to a calculated path in the virtual terrain. In doing so, this pathfinding algorithm considers the terrain scenario and the particularities of each kind of vehicle being simulated (such as size, weight, capacity of making sharp turns, etc.). In this task, the overall aim is to park each vehicle inside the waiting position (represented on Figures 2 (D) and 2 (E)).
Figure 2 - (a) Example of a battery of military vehicles arriving at the waiting position. (b) Battery units entering the waiting position in a line convoy formation. (c) 2D representation of the battery units moving as a CGF aggregate. (d) 2D representation of the battery units disaggregated since each unit needs to park in its chosen position. (e) Military vehicles parked inside the waiting position
This task-oriented and parameterized navigation algorithm also deals with unexpected situations in the simulation exercises, which can be situations inserted by simulation instructors. For example, the battery units move as a convoy to the waiting position (Step 3). When all units successfully arrive at this location, the overall convoy task is complete with success. This allows the convoy to initiate the next subtask in Step 4. In addition to other simulation events, however, there is a possibility of a battery unit breaking due to a variety of problems during this convoy navigation. It means that even when a particular vehicle of the battery convoy does not complete its subtask, the overall convoy navigation algorithm needs to detect this situation and complete its original navigation task. To organize these movement actions, the problem situation of a broken unit is treated by using internal state control settings implemented in each battery vehicle. When the overall semi-autonomous navigation algorithm detects such problems, simulation users are requested to provide a solution for them. Otherwise, the autonomous behavior of the task-oriented algorithms is programmed to complete such navigation task without considering the broken unit. 4.2 An autonomy control solution for virtual tactical simulations As stated in1, it is easier to insert a human-in-the-loop component in an already existing autonomous simulation system than vice-versa. In this project, this same approach is followed since fully autonomous functionalities for CGFs were first implemented and tested in the proposed navigation framework. After that, semi-autonomous capabilities were gradually inserted in the CGF navigation algorithms in order to consider inputs from both users and other simulation systems involved in the distributed simulation architecture. One simple example of such blended autonomy implemented is related to the simulation situation in which one battery vehicle belonging to a simulated convoy steers according to the manual input of users which are interacting with the simulator. Even though this vehicle is being controlled by human users, other battery vehicles in this convoy still follow an autonomous behavior. In effect, these autonomous CGFs navigate according to the convoy formation movements chosen in the parameterized algorithms.
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
541
In this case, the overall steering behaviors of these autonomous battery vehicles use a follow-the-leader strategy in which this leader vehicle is the user controlled vehicle. Choice regarding the movement strategy to be used and the unit which has the convoy leader role are also related to parameters that are set up in the algorithms. In effect, this simple example highlights the importance of the different levels of autonomy for the CGFs involved in the simulations as this importance is related to the direct user interference and interaction with the task-oriented navigation algorithms used by the simulation system. 4.3 A global and local navigation solution for virtual tactical simulations In the studied simulation system, roads available in the virtual terrain are represented by a graph structure. Based on this graph, a standard A* algorithm is used in the computation of a navigation path between two points belonging to these roads. Despite such representation for roads, this virtual scenario does not have pre-processed representations for terrain areas that do not have roads since the content of these areas can change while the simulations are running according to inputs from simulation instructors. For example, areas can become more or less wet as to introduce simulation obstacles to be resolved by military trainees. In general, if the battery vehicles need to navigate across terrain regions with no roads, navigation information throughout these regions has to be processed during simulation execution time. Due to the large size of the virtual terrain, a Quadtree structure is used to represent theses areas of open fields, forests, wet/dry fields, etc. As to optimize the execution of the pathfinding search over these areas, an A* algorithm hierarchically traverses this Quadtree structure. After quickly eliminating regions that contain obstacles as represented in higher nodes of the Quadtree structure, the pathfinding algorithm returns a list of nodes indicating the path that the battery vehicles should use when the search reaches the deepest level of the Quadtree. Having such path in the virtual terrain, the navigation algorithm also executes a local steering behavior algorithm whenever an unforeseen static or a dynamic obstacle is found (e.g. another vehicle in movement inside that local region). Both the global pathfinding and the local steering behavior are fundamental since the single use of a reactive steering-based solution to this problem is not enough to realistically simulate this domain/task-specific situation. 4.4 A convoy coordinated movement solution for virtual tactical simulations The development of tactical military simulation tasks often involves a group of heterogeneous units following different kinds of convoy formation movements in the virtual simulation scenario. In the tactical simulations of an artillery battery, this CGF group maintains formation while moving between different positions as to complete the task regarding the “occupation of a waiting position”, for instance. In doing so, local steering force-based algorithms enable these CGFs to fulfill cohesion requirements between the individual elements that constitute the battery convoy. As these units are displaced in different kinds of battle positions, they are also strategically organized in these chosen locations, where regions and obstacles in this virtual simulation environment are either previously pre-processed or not by the simulation system. Battery vehicles also follow a military doctrine, in which alternative military formations are used in particular artillery deployment tasks. When following the doctrine, lightweight vehicles may not be allowed to speed up in front of the group, arriving at a certain destination first, while letting the heavier units of the battery convoy behind without protection. In practice, the roles that the battery vehicles have while following a selected formation are set up by parameters in the task-oriented navigation algorithms used. In contrast, the underlying control algorithms also allow the battery units to “break” formations in specific situations. These CGFs leave the convoy cohesion if they detect or suffer an enemy attack, for example. Most importantly, they also break formation as to execute specific simulation tasks requested by users or other simulation systems. 5. Related works Blended simulation architectures involve the integration of constructive systems with human-in-the-loop capabilities as applied in, for instance, the Royal Australian Air Force (RAAF) simulation system 1, and the intelligent automated agents for large-scale, realistic simulation environment called SAFORS (semi-automated forces) in the SIMNET environment4. These architectures lead to semi-autonomous systems that combine a constructive navigation behavior with the possibility of human interference as approached in our framework. Through simulations developed
542
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
when our framework architecture is used, users can interact with a simulation exercise as to make sense of how the military doctrine works, permitting them to review the decision-making process to recognize and avoid possible errors. Semi-autonomous navigation is a recurrent topic in the research field of robotics, and mainly in the field of autonomous driving cars. In19, a DoD`s Demo III UGV (mobile robot) program is described as a testbed for autonomous navigation research. The goal is the creation of a UGV capable of autonomous and semi-autonomous operations as part of mixed military exercises, which are issues also approached by our proposed framework. Similar to such UGV solution, our CGF implementations are also able to explore both deliberative and reactive navigation behaviors as their combination provides a solution for global and local navigation problems. In another work 20, a planning algorithm is used to determine a path which is assumed to be the safest path or the best-path case to be followed by an agent representing a vehicle. At the same time, this vehicle senses and predicts a threat level based on real vehicle navigation situations. If this threat level surpasses a limit, the autonomous driving system assumes the vehicle control; otherwise, a semi-autonomous control is left in place. A similar system containing both path planner and thread prediction components is presented in21, which is a similar solution to our hybrid framework proposal. In that work, the vehicle also has a local navigation component which is responsible to quick maneuvering the vehicle through the path returned from the planner. As these semi-autonomous works are focused on the field of robotics, however, they rarely consider the task-oriented doctrine-based features that are fundamental in the implementation of virtual learning environments for the development of realistic military simulations. Moreover, their applicability and impact in other areas such as simulation systems ought to be examined further, where steps in this direction are detailed in the proposed framework. In contrast to semi-autonomy based techniques, hybrid navigation techniques have been explored in the field of computer games, which are at a certain extent relevant to the development of simulation systems. As described in22, a hybrid pathfinding approach is used in the popular RTS game Starcraft as to achieve navigation in a terrain with static and dynamic obstacles. There, the A* algorithm is used to calculate a path if there are no obstacles in the field of view of the game agent under consideration. If the agent encounters a dynamic obstacle, two different techniques are explored: potential fields and flocking algorithms5. In effect, such kinds of computer game works demonstrate that both hybrid techniques out-perform a non-hybrid system. However, the computer game solutions are commonly applied to terrain representations that are very small and not much complex if compared to realistic terrain representations which are primarily required by many real-life simulation systems. Besides, the potential field hybrid solutions are often related to a high computational cost as far as the size and complexity of the terrain and the number of agents navigating on it are concerned. In this context, techniques to represent large terrains and achieve navigation on them are explored in23, motivating the development of the global navigation technique used in our framework. In this case, a technique that combines a terrain representation in a Quadtree data structure with the A* algorithm, creating the Quad* algorithm is proposed in24. Besides not approaching large and complex virtual environments, such proposals do not consider an overall framework solution comprising the interaction of the terrain models with navigation techniques and switching modes of autonomy, as jointly addressed in our framework. The proposed framework also differs from those presented in computer games as it manipulates the large terrain characteristics in a dynamic way according to the doctrine-based navigation performed by CGFs. In conclusion, despite the existence of partial solutions for the different problems tackled by the proposed framework, there is a lack of adjusted solutions to handle these problems together in a complex simulation environment, as the proposed framework does. In light of this fact, the framework presents a step forward in the direction to providing a new solution for the development of realistic simulations focusing the training of domain users. 6. Final remarks This paper describes a navigation framework that addresses task-oriented (semi) autonomous navigation behavior implementations for virtual tactical military simulation exercises. This framework deals with local and global navigation issues, the treatment and avoidance of static and dynamic obstacles, the existence of heterogeneous military units along with their particular navigation behaviors, and the different degrees of (semi) autonomy and user interaction in the underlying distributed simulation architecture. In effect, the task-oriented and parameterized implementations amount to a solution that maintains a degree of generality which makes it possible to reuse the CGF algorithms in different simulation applications. In addition, the combination of local and global navigation solutions techniques with
Juliana R. Brondani et al. / Procedia Computer Science 112 (2017) 534–543 Juliana Brondani et al/ Procedia Computer Science 00 (2017) 000–000
543
adequate data structures for virtual terrain representation allows the pathfinding in such terrains, while maintaining a realistic local steering behavior for the simulated CGFs. The capability of switching autonomy also creates a better training experience as simulation system users are able to influence the decision-making process of the simulation system. Finally, this work offers the detailed discussion of a concrete military simulation problem as to provide evidence for the applicability of the CGF navigation characteristics now properly organized in the presented framework. To enhance this framework, future works will be directed to the exploration of different levels of semi-autonomy in the simulations, and how they affect the simulation system executions. Acknowledgements We thank the Brazilian Army for the financial support through the SIS-ASTROS Project (813782/2014), developed in the context of the PEE ASTROS 2020. References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24.
Heinze, C., Goss, S., Josefsson, T., Bennett, K., Waugh, S., Lloyd, I., et al. Interchanging agents and humans in military simulation. AI Magazine 2002;23(2):37. M.J., Brown, M.S. Software engineering a multi-layer and scalable autonomous forces ai for professional military training. In: Proceedings of the 2016 Winter Simulation Conference. IEEE Press; 2016, p. 3122–3133. Ko, S.G., Yun, W.S., Lee, T.E. Modeling and simulation based analysis of effectiveness of tactical level chemical defense operations. In: Proceedings of the 2016 Winter Simulation Conference. IEEE Press; 2016, p. 3098–3109. Tambe, M., Johnson, W.L., Jones, R.M., Koss, F., Laird, J.E., Rosenbloom, P.S., et al. Intelligent agents for interactive simulation environments. AI magazine 1995;16(1):15. Kapadia, M., Badler, N.I. Navigation and steering for autonomous virtual humans. Wiley Interdisciplinary Reviews: Cognitive Science 2013;4(3):263–272. Page, E.H., Smith, R. Introduction to military training simulation: a guide for discrete event simulationists. In: Simulation Conference Proceedings, 1998. Winter; vol. 1. IEEE; 1998, p. 53–60. Stevens, J., Kincaid, P., Sottilare, R. Visual modality research in virtual and mixed reality simulation. The Journal of Defense Modeling and Simulation 2015;12(4):519–537. SAE International Surface Vehicle Recommended Practice, “Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles”, SAE Standard J3016, Rev. Sept. 2016 Woodyard, C. 2010. GM takes its EN-V two-wheeler out for a spin. USA Today. Oh, K.K., Park, M.C., Ahn, H.S. A survey of multi-agent formation control. Automatica 2015; 53:424–440. Tolk, A. Engineering principles of combat modeling and distributed simulation. John Wiley & Sons; 2012. Studer, R., Benjamins, V.R., Fensel, D. Knowledge engineering: principles and methods. Data & knowledge engineering 1998;25(1- 2):161– 197. Evertsz, R., Thangarajah, J., Yadav, N., Ly, T.C. Agent oriented modelling of tactical decision making. In: Proceedings of the 2015 International Conference on Autonomous Agents and Multiagent Systems. International Foundation for Autonomous Agents and Multiagent Systems; 2015, p. 1051–1060. Zilberstein, S. Building strong semi-autonomous systems. In: AAAI. 2015, p. 4088–4092. De Berg, M., Van Kreveld, M., Overmars, M., Schwarzkopf, O.C. Computational geometry. In: Computational geometry. Springer; 2000, p. 1–17. Snook, G. Simplified 3d movement and pathfinding using navigation meshes. Game Programming Gems 2000;1(1):288–304 Pottinger, D. Implementing coordinated movement. Game Developer Magazine 1999;48–58. Reynolds, C.W. Flocks, herds and schools: A distributed behavioral model. ACM SIGGRAPH computer graphics 1987;21(4):25–34. Shoemaker, C.M., Bornstein, J.A. The demo iii ugv program: A testbed for autonomous navigation research. In: Intelligent Control (ISIC), 1998. Held jointly with IEEE International Symposium on Computational Intelligence in Robotics and Automation (CIRA), Intelligent Systems and Semiotics (ISAS), Proceedings. IEEE; 1998, p. 644–651. Anderson, S.J., Peters, S.C., Pilutti, T.E., Iagnemma, K. An optimal-control-based framework for trajectory planning, threat assessment, and semi-autonomous control of passenger vehicles in hazard avoidance scenarios. International Journal of Vehicle Autonomous Systems 2010;8(2-4):190–216 Gray, A., Gao, Y., Lin, T., Hedrick, J.K., Tseng, H.E., Borrelli, F. Predictive control for agile semi-autonomous ground vehicles using motion primitives. In: American Control Conference (ACC), 2012. IEEE; 2012, p. 4239–4244. Hagelbäck, J. Hybrid pathfinding in starcraft. IEEE Transactions on Computational Intelligence and AI in games 2016;8(4):319–324. Algfoor, Z.A., Sunar, M.S., Kolivand, H. A comprehensive study on pathfinding techniques for robotics and video games. International Journal of Computer Games Technology 2015:7. Dooms, A. Parallel multi-agent path planning in dynamic environments for real-time applications. MA Thesis, Ghent University, Faculty of Engineering and Architecture, 2013;
