Year: 2018

Authors: Benjamin Bell, Ph.D., Winston Bennett, Jr., Ph.D., William J. Clancey, Ph.D.

Abstract: Today’s warfighters are being trained for information-rich, networked, automated battlespaces. But what happens when information access is disrupted? How can training prepare personnel to win in contested environments with austere access to sensors, navigation and communications? A renewed focus across DoD on near-peer adversaries is highlighting the need to answer these questions by incorporating AntiAccess/Area Denial (A2AD) effects (e.g., datalink jamming, GPS spoofing) into training. Despite continuing improvements in simulations, modeling how people and technology (a “sociotechnical system”) coordinate under nominal and denied conditions requires new approaches. Simulations must, for instance, model the disruptive effects of communications degradations on mission effectiveness. We are exploring an area of relevance across the training community, simulating sociotechnical processes to train today’s forces for denied environments. AFRL, Eduworks Corporation, and the Florida Institute for Human and Machine Cognition are exploring pilot training for contested environments, focusing initially on A2AD effects in denied airspace. We use a Government-owned framework called Brahms to model agents, objects, geography, cultural features and information systems. For interoperability with existing training environments, we employ an AFRL tool for connecting models to simulations using Distributed Interactive Simulation, m2DIS, enabling Brahms models to serve as constructive agent controllers. The testbed includes visual, drag-and-drop scenario generation and automated visualization that extracts patterns and trends from multiple scenario runs under different initial assumptions. We discuss how the Brahms Contested Airspace Simulation Testbed (Brahms-CAST) will enable simulations to incorporate A2AD effects and support experimentation and analysis of contested environments.

Year: 2018

Author: Ran Shneor

Abstract: The Israeli Air Force (IAF) recently entered the era of fifth generation aircraft. Israel was the first country outside the United States to receive, operate, and maintain the F-35A Lightning II aircraft. F-35 are maintained by young technicians (ages 18-20) during their compulsory military service after a short maintenance course. Many of them lack aerial experience or have no technical background at all. Teaching inexperienced personnel about complex systems, like the F-35, in a constantly changing environment is a significant training challenge for the IAF. To overcome this challenge, the IAF employed innovative approaches to technical training. The focus of this innovative approach was on adjustments to the system (i.e. the organization) to efficiently and effectively meet individual training needs. This paper presents unclassified and non-commercial information of the different activities in technical training, followed by a detailed description of the innovative actions taken in preparation for the F-35 reception in the IAF. It is based on the Technological Pedagogical And Content Knowledge (TPACK) model. An important insight based on the IAF's experience implies innovative technical training is achieved by contradiction and interaction between opposite forces and perceptions. The effectiveness measurements of the IAF’s performance show remarkable results. Guidelines are deduced based on the IAF's experience. These guidelines are relevant for organizations facing a leap in technology or major changes in operations and training environments.

Year: 2018

Authors: Randy Jensen, Sowmya Ramachandran

Abstract: Many modern operational performance environments produce significant data artifacts that collectively constitute rich libraries of decision-making examples. For domains where expert decisions are guided by constraints, there is the potential to automatically derive the constraints themselves from expert performance data. This paper discusses a datadriven machine learning approach to modeling constraints, implemented in an authoring tool coupled with a simulation-based training environment for satellite planners. In this domain, the planner’s task is to create a 7-day schedule of requested satellite contacts, while meeting a range of specialized planning constraints which vary for different satellites with different missions. The training goal is to assess planners’ decisions in simulation-based scenarios and provide feedback, which requires automated performance assessment measures with knowledge of planning constraints. For this application, the authoring tool provides a utility to directly process operational source data, in this case consisting of archived records of satellite requests from previous periods. This produces derived constraints, which authors then review, edit, and annotate as needed before linking the constraints to runtime assessment mechanisms for exercises. Beyond the initial focus on generating automated assessment with this datadriven approach, the development process uncovered other useful applications for the ability to derive constraints from operational data. For example, one phase of the satellite planning process involves deconflicting one’s own satellite support requests from those involving other satellites that may be seeking to simultaneously use the same resources, such as a specific ground antenna. In order to support individual training for this task, an automated agent was created to produce realistic simulated conflicts with the planner’s requests, based on constraints mined from operational data. This research also helped to uncover drawbacks in the data-driven approach for some domains, so this paper discusses applicability and limitations in more general cases beyond the initial satellite planning application.

Year: 2018

Authors: Ronald G. Moore, Matthew J. Reilly, Tony Pelham

Abstract: Automated collection-to-construction of terrain databases is a critical capability envisioned for future U.S. Army training systems. The challenge is how to automatically produce terrain data that supports both visual rendering and simulated reasoning with content sufficient to train ground forces in dense urban environments. The process of automated terrain construction begins with surface capture. Drones and ground-robots are deployed, capturing large amounts of raw surface data. Processing the surface data yields point clouds or 3D polygonal meshes, providing an initial 3D terrain model, typically with very high point/polygon densities and large raster memory requirements. While certain applications may be able to utilize these terrain models directly, most visualization applications, require additional processing to generate well-formed model geometry, sharp textures, door and window apertures, and material classifications. This additional processing, performed on the point cloud or 3D polygonal mesh, extracts point, line, and polygon feature geometries along with descriptive feature attributes (e.g., height, roofline, roof-type). A bare earth elevation model is generated to provide a ground surface in which to place the extracted 3D features. The final enabler of the terrain construction process is the automated generation of 3D models from the feature and attributed data. This paper reports on research which expands automated extraction of attributes from images through deep-learning and image processing techniques, identifying structural dimensions, apertures (e.g., doors, windows), appendages (e.g., A/C-Unit, chimneys), colors, and materials. From this set of enhanced attributes, geo-representative 3D models are procedurally generated. In addition, from the same set of enhanced attributes, geo-representative building-interiors are speculated and procedurally generated. This paper details these image processing and deeplearning techniques, describes the enhanced feature attributes that are extracted, explains the methods for interior speculation, and details the techniques for procedural 3D model generation. The paper provides lessons-learned and recommends a new standard for procedural model generation.

Year: 2018

Author: Tyler Ricks

Abstract: Many modern simulations struggle to implement human and animal characters into their solutions, due to the challenges of rigging and animation that differ greatly between species. Multiple obstacles challenge implementation, including sorting out different world spaces, integrating disparate toolsets, and addressing design requirements. Certain solutions, such as motion capture, are sufficient for most needs for humans, but are almost useless for certain animals. Although some video game solutions offer support for animal characters, differences between the gaming and simulation industries are often so great that using third-party software solutions in simulation environments is expensive, time consuming, and removes the ability to control animations within the simulation runtime. This paper describes how an animation system was created that allows for human and quadruped characters to the built for simulation. The solution allows for input of animation data from game resources, motion capture, or keyframe animation techniques common in the animation industry. This process is enhanced by the fact that it allows for the full use of available modeling, rigging, and animation toolsets. This is important for the simulation community because it grants full access to a wide variety of animation resources, greatly lowering the cost of implementing many different species into existing or future databases. This paper describes the features and advantages of such a system, which allows for direct control of animated humans and animals in the simulation environment using standard simulation controls, as well as offering host control of the characters directly. It compares the processing load put on systems using older systems and techniques. Finally, future potential developments and possibilities are explored.

Year: 2018

Authors: Rick D. Giovengo, Ph.D, George Buck

Abstract: Objective: The Federal Law Enforcement Training Centers (FLETC) initiated research to determine if a blended approach of instruction was effective for teaching the driving technique “determining and following an ideal line of travel (LOT).” The LOT is defined as the most efficient path to steer a vehicle when traveling through a turn. It is commonly taught in emergency vehicle operations training. Currently, this concept is taught using a lecture block and a hands-on driving on an outdoor driving range. The problem is students do not get as much driving time due to the limited amount of cars, instructors, and driving ranges available. Research Questions: Can participants who receive training in a blended learning environment perform better on the driving range practical examination? Can participants who receive training with a blended and traditional learning environment transfer that knowledge when presented with a new environment? Can participants who receive training in a blended learning environment (both simulated and live driving) gain a better understanding of the principles of LOT on a cognitive test? Methods: Participants were given a one-hour lecture on the concepts of an ideal LOT. The group was divided into two. The control group went to the driving range to practice LOT techniques. The experimental group was taken to the driving simulators to practice LOT techniques. After practice, the experimental group was then taken to the driving range and allowed to practice with the other group. At the end of practice both groups were giving a driving assessment on the familiar range and a novel range. Then both groups were given a written exam. Results: The differences in driving performance and passing rate between groups were minimal, although a statistical significance was found between groups on the familiar range. Lack of training transfer was found in both groups when taken to the novel range. The blended / simulator group scored higher on the written exam. Conclusion: Use of simulators in a blended approach produced similar results compared to traditional instruction for LOT.

Year: 2018

Authors: Vikram Mittal, Robert Kewley, Brett Lindberg

Abstract: A significant challenge in teaching simulation to undergraduate students is to find a way to allow them to model a real world referent system within time and student skill constraints. Several research sources highlight not only the important challenge of model development (Garcia and Ceneno, 2009, Tako, 2011) but also the increased need for model development instruction among engineers (Grasas et. al., 2013, Saltzman and Roeder, 2013). One approach to this challenge is to use a general purpose discrete event simulation software package within the course, but this presents two challenges. Teaching the package to the students takes significant time, and the package introduces limitations which may restrict their ability to model certain realworld referents, particularly in the engineering domain. A conceptual approach to solving this problem is to use a model development paradigm that abstracts away the interface to the simulation infrastructure while still allowing the students to use the full expressive nature of a programming language. Two undergraduate courses at the United States Military Academy employed this strategy via the Discrete Events Specification System – Distributed Modeling Framework (DEVS-DMF) (Kewley et. al, 2016). The DEVS abstraction allowed students to think about their model as a simple state change function with defined inputs and outputs, and DMF allowed them to program in a cloud-based Jupyter Notebook using the Python language. Students in a combat modeling course employed a variety of models to understand drone jamming, and students in an engineering capstone project employed models to account for human factors in rifle marksmanship. The effectiveness of this approach was assessed through student grades, exit-interviews, and course-end surveys. These assessments showed an increased understanding of the model development process, and students also reported greater ownership of their models. However, this experiment also highlighted some weaknesses in their understanding of underlying methodologies and programming skills.

Year: 2018

Author: MAJ Gareth W. Collier (retd)

Abstract: Live field firing represents the apex of operational preparedness training for land forces, yet carries persistently high risks of dangerous occurrences and fratricide. To date simulation has developed various individual tools to aide in the preparation, planning, conduct, optimization and analysis of live field firing, yet the collective application of these to systemically enhance training outcomes and reduce risk has not materialized. This paper proposes employing multiple simulation technologies to support essential military live field firing within acceptable risk thresholds, whilst reducing the ongoing instances of fratricide modern armies continue to experience. Currently live field firing planning is largely paper based. Manual creation of Range Danger Areas onto clear overlays are positioned on paper maps of the training area. A supervisor then manually reviews these range templates, applying several hundred pages of doctrine from memory to verify safety. Complexity and risk both increase exponentially for more intricate range practices incorporating multiple phases, combined arms effects, simultaneous manoeuvre, night activities, lasers and joint assets. For such practices, several thousand pages of complex safety doctrine must be applied, with sequencing or policy errors resulting in increased risk to personnel during live conduct. Simulation is conspicuously absent from this process, despite its potential to deliver relevant tools in planning and conduct phases to reduce risk and enhance capability. This paper proposes integrating various simulation technologies to persistently reduce risk in military live fire training. In the planning phase this includes geospecific replication of military live fire ranges, automated construction / visualization tools for range design, creation and sequencing, and automated rule checking of doctrine safety and compliance. During the conduct phase this enables the replication of terrain / range templates with real-time geolocated troops, platforms and targets to anticipate emergent risks – the core reasons fratricide occurs – then prompt commanders to mitigate these risks before they eventuate.

Year: 2018

Authors: Julia L. Walsh, Kent C. Halverson, Eric Watz, David Malek

Abstract: The Department of Defense faces increasingly complex mission sets, demanding unprecedented warfighter performance and flexibility in unpredictable situations. Autonomous intelligent agent (AIA) technology promises to significantly relieve these demands, in both training and operational settings. In the future, warfighters and AIAs will work jointly as part of mixed human-machine teaming (HMT) environments, wherein AIAs will routinely assess warfighter cognitive states to optimize learning, task management, feedback delivery, and performance. To meet the desired future state, training must equip humans with HMT skills necessary to optimize this collaboration. Learning environments must allow humans to experience authentic scenarios with HMT interactions that can be measured and assessed to provide real-time and after-action feedback. To ensure HMT skill acquisition and retention, the AIA should meet the following system requirements. It needs to operate autonomously in complex environments, be adaptable, and be able to enhance learning by delivering strategic workload-based personalized feedback. One way of achieving the aforementioned goals is by engineering a testbed that would meet the said system requirements and would train operators to interact with AIA and human teammates. The envisioned testbed would represent a fully deployable live, virtual, and constructive training system featuring integrated inner-loop (during the training scenario) and outer-loop feedback (after the scenario) capabilities along with a multi-modal measurement suite. The testbed would serve dual purpose – one for realizing operational training needs and the other for conducting within-subject multi-trial scenario-based studies to examine whether or not the presence of AIA impacts warfighter performance. Consistent with a growing body of literature which suggests that human performance tends to improve with an addition of an AIA (e.g., Mercado et al., 2016; McKendrick et al., 2014), we predict that warfighter performance will improve when AIA is present. While the research potential of the testbed is limitless, we offer several future research directions in this paper.

Year: 2018

Authors: Kevin Hulme, Aaron Estes, Matthias Schmid, Emmanuel Gil Torres

Abstract: Modeling & Simulation (M&S) is an effective mechanism for bridging the gap between experimentation and implementation; between basic theory, and real-world application. For example, in the field of engineering systems and dynamics, trainees can leverage M&S to experience vehicle control within a high-fidelity game-based environment to simulate the operation of a physical vehicle. Engineering insights gained through “active” learning with simulation are essential; as ground vehicle transportation continues towards fully self-driven vehicles, there remains numerous technological and human factors challenges yet to be overcome. This paper discusses the experimental design of a game-based simulation environment implemented for a road vehicle dynamics (RVD) university-based engineering curriculum. The Simulation exercise is The (ISO 3888-2) Moose Test - an evasive vehicle maneuver, typically performed by on-road experts, to determine the thresholds of vehicle handling under aggressive driving conditions (e.g., swerving to dodge a moving obstacle). Our developmental basis is to provide a virtual-constructive Simulation environment for novice learners to: a) experience this maneuver under controlled conditions (e.g., two courses; three entry speeds; with or without electronic stability control), b) optimize performance (i.e., maintain vehicle control and remain path-centric), and c) obtain an improved understanding of vehicle stability that would not be feasible on an actual proving grounds. Using both Simulator and self-report metrics, we will further observe if experimental performance correlates to tendencies relevant to driving and learning, which could inform the degree to which Simulation-based training is better suited towards certain categories of drivers. As critical Broader Impacts, this insight could advise how the operation of next-generation vehicles (i.e., mechanisms for operator intervention) can be tailored to individual differences (e.g., age, experience, aggressive tendencies) in specific driver types. Likewise, this M&S implementation has extensibility to military applications (e.g., pilots for aircraft) within transportation and human factors research.