A key obstacle for training an autonomous agent in real air-to-air combat is the lack of available training data, which makes it difficult to apply supervised learning techniques. Self-play is a method that can be used where an agent trains against itself or against versions of itself without imitation data or human instruction. Agents training only against themselves learn brittle strategies that do not generalize very well, which is why training against a distribution of strategies is necessary to improve robustness. In this thesis, we study two problems. First, what is a robust strategy, and how do we evaluate it? Secondly, how do we increase the robustness of strategies learned in a self-play setting by adapting the training distribution? The problems are significant to study because self-play is a very promising method of training not only for air combat but in any non-cooperative problem setting where a simulator can be used to gather training data with no human in the loop. Specifically, in the aircraft industry, the cost of gathering samples is incredibly high. To evaluate the robustness of a population of strategies we turned to evolutionary game theory and connected theα-rank algorithm to what we perceive as robustness. Theα-rank induces a strict ordering over the set, which we then take as an evaluation of the robustness of the strategies. We validated that a highα-rank correlated well with performing well in an out of population evaluation. To study how the robustness of a population correlated with training distributions, we trained populations against four different training distributions. We used the uniform, δ-uniform, andα-rank distributions that rely on no information, information on the training process, and information on the robustness of agents, respectively. We also designed a novel amortizedα-rank training distribution that combines the information behind the δ-uniform and α-rank distributions, and we showed that it induced superior robustness properties in the learned strategies. Our efforts indicate that even better training distributions can be produced, which is useful when looking at using self-play in the future. / Ett stort hinder vid träning av en autonom agent för riktig luftstrid är bristen på träningsdata vilket gör det svårt att applicera övervakad inlärning. Självspelande är en metod där agenter tränar mot sig själva eller mot versioner av sig själva utan tillgång till data att imitera och utan mänsklig instruktion. Agenter som tränar enbart mot sig själva lär sig dock sköra strategier som inte generaliserar tillräckligt väl. Därför krävs träning mot en distribution av strategier för att en agent ska kunna bli robust. I denna uppsats studerar vi två problem. Ett, vad är en robust strategi och hur utvärderar vi den? Två, hur ökar vi robustheten hos strategier som tränats fram via självspelande genom att anpassa träningsdistributionen? Problemen är relevanta att studera på grund av att självspelande är en lovande metod som kan användas inte bara för luftstrid utan för ett brett spektrum av icke-kooperativa problemområden där en simulator kan användas för att genomföra datainsamling utan en människa i loopen. Speciellt inom flygplansindustrin är kostnaden för insamling av riktig träningsdata extremt hög. För att utvärdera robustheten i en population av strategier vände vi oss tillevolutionär spelteori och koppladeα-rank-algoritmen till vad vi uppfattar som robusthet.α-rank, som vi tolkar som en utvärdering av robusthet, induceraren strikt ranking över en mängd av strategier. Vi validerade att en högα-rankkorrelerade väl med goda resultat vid en utvärdering av strategierna i en annan population. För att studera hur robustheten i en population korrelerade med användandet av olika träningsdistributioner tränade vi populationer mot fyra olika tränings-distributioner. Vi använde den uniforma distributionen, den δ-uniforma distributionen ochα-rank-distributionen, som baseras på ingen information, information om träningsprocessen respektive information om agenternas robusthet. Videsignade även en amorteradα-rank-distribution, som kombinerar informationen bakom den δ-uniforma distributionen ochα-rank-distributionen, och vi visade att träning mot den nya distributionen resulterade i mer robusta strategier. Våra resultat indikerar att det är möjligt att skräddarsy ännu bättre träningsdistributioner, vilket är användbart när vi tittar på att utnyttja själv-spelande i högre grad i framtiden.

The goal of this work has been to develop a design suggestion of a visualization system that supports training instructors and air force units during After Action Reviews (AAR) in training at the Swedish Air Force Combat Simulation Center (FLSC). The suggestion was developed through a design study consisting of interviews, a demonstration and a design workshop in parallel with an iterative prototyping process. The study resulted in a visualization system that indicates enemy aircraft at threatening distances to support identification of critical situations. To support the following threat assessment, visualizations that automatically present radar lock-on information and missile envelopes are suggested. Although radar lock-on information and missile envelopes has been presented in earlier visualizations they have not been adapted to the mode of operations of the AAR. Adapting the visualizations to the AAR would make the threat assessment easier and more effective to conduct. The purpose of the visualization is to make it easier for training instructors and air force units with limited experience of simulation training to conduct a high quality AAR through emphasizing relevant situations and points of interest.

International Telemetering Conference Proceedings / October 25-28, 1999 / Riviera Hotel and Convention Center, Las Vegas, Nevada / This paper provides a current review of a new installed system test facility (ISTF) capability for the Air Force and Navy. The requirements, design characteristics, and status of the joint-service Communications, Navigation, Identification Simulator (CNIS) developments will be covered along with their relationships with the Air Force’s Avionics Test and Integration Complex (ATIC) and the Navy’s Air Combat Environment Test and Evaluation Facility (ACETEF) ISTFs. These developments provide the services an interactive spatially, temporally, and tactically coherent signal environment for development and operational test and evaluation. The Joint Communications Simulator (JCS) and Joint Data Link Simulator (JDLS) capabilities, integration aspects, and development schedules (2000 IOC) will also be addressed. Finally, installed system test and evaluation concepts, both Air Force and Navy, using the simulators will be previewed to assist upcoming development programs in identifying potential applications.

A study of communication content, frequency and problems was performed at the Swedish Air Force Air Combat Simulation Centre, FLSC. The purpose was to investigate the connection between team communication and performance variables, to study communication problems during air combat and to employ and develop the PILOT method of communication problem analysis. Ten fighter pilots and four fighter controllers engaged in simulated air combat beyond visual range in teams on two sides, Blue and Red, with four pilots and one fighter controller each. Eight such training runs lasting 15 to 25 minutes were analyzed with regard to the Blue team’s communication during launch situations, i.e. one minute before launching a missile to either missile impact of failure, and problems during whole runs. The performance of the Blue team was rated by an experienced training instructor in six of the runs. It was hypothesized that positive outcomes would be associated with high communication frequencies but low communication problem rates. Results showed that high utterance frequencies were positively associated with positive global outcomes, i.e. wins, but not associated with high instructor ratings or launching hitting missiles. Incoherent results were found for communication problems. The results were discussed and further research suggested.

Studiens syfte är att öka förståelsen för möjligheterna med modern artificiell intelligens (AI) vid militär användning genom att bidra med information om ny teknik. Moderna civila applikationer av datorseende som skapats genom användande av så kallade artificiella neurala nät visar resultat som närmar sig den mänskliga synens nivå när det gäller att känna igen olika saker i sin omgivning. Denna studie motiveras av dessa observationer inom området AI i förhållande till situationer i luftstrid då pilotens syn används för att känna igen flygplan innan det bekämpas. Exempelvis vid användande av hjälmsikte riktar pilotens ögon målsökaren hos en luftmålsrobot mot det flygplan som robotens målsökare sedan låser på. Utanför visuella avstånd kan pilotens ögon av naturliga skäl inte användas på detta sätt, varför datorseende använt i en luftmålsrobot undersöks. Resultaten från studien stödjer att datorseende genom användande av artificiella neurala nät kan användas i en luftmålsrobot samt att datorseende kan utföra uppgiften att känna igen stridsflygplan. / This study is aimed at increasing the knowledge to those concerned within the Armed Forces by providing information on the possibilities of modern artificial intelligence (AI). Motivation comes from observations of civilian technology on the use of AI in the field of Computer Vision showing performance equaling the level of the human vision when using the technology of Deep Learning of Artificial Neural Nets. In air-combat aircraft the pilot´s vision is used for recognizing the aircraft that is about to be shot down. For example when utilizing helmet mounted displays, the seeker of an air-target-missile is directed upon the aircraft on which the pilot´s eyes are looking. However when air-target-missiles are utilized beyond visual range the pilot´s vision cannot help in directing the seeker on a specific target. Therefore computer vision within an air-target-missile is studied. The results of the study support that the technology of neural networks may be used in an air-target-missile and that computer vision provided by this technology can do the job of recognizing a combat aircraft.

3D user interface technologies have the potential to make games more immersive & engaging and thus potentially provide a better user experience to gamers. Although 3D user interface technologies are available for games, it is still unclear how their usage affects game play and if there are any user performance benefits. A systematic study of these technologies in game environments is required to understand how game play is affected and how we can optimize the usage in order to achieve better game play experience. This dissertation seeks to improve the gaming experience by exploring several 3DUI technologies. In this work, we focused on stereoscopic 3D viewing (to improve viewing experience) coupled with motion based control, head tracking (to make games more engaging), and faster gesture based menu selection (to reduce cognitive burden associated with menu interaction while playing). We first studied each of these technologies in isolation to understand their benefits for games. We present the results of our experiments to evaluate benefits of stereoscopic 3D (when coupled with motion based control) and head tracking in games. We discuss the reasons behind these findings and provide recommendations for game designers who want to make use of these technologies to enhance gaming experiences. We also present the results of our experiments with finger-based menu selection techniques with an aim to find out the fastest technique. Based on these findings, we custom designed an air-combat game prototype which simultaneously uses stereoscopic 3D, head tracking, and finger-count shortcuts to prove that these technologies could be useful for games if the game is designed with these technologies in mind. Additionally, to enhance depth discrimination and minimize visual discomfort, the game dynamically optimizes stereoscopic 3D parameters (convergence and separation) based on the user's look direction. We conducted a within subjects experiment where we examined performance data and self-reported data on users perception of the game. Our results indicate that participants performed significantly better when all the 3DUI technologies (stereoscopic 3D, head-tracking and finger-count gestures) were available simultaneously with head tracking as a dominant factor. We explore the individual contribution of each of these technologies to the overall gaming experience and discuss the reasons behind our findings. Our experiments indicate that 3D user interface technologies could make gaming experience better if used effectively. The games must be designed to make use of the 3D user interface technologies available in order to provide a better gaming experience to the user. We explored a few technologies as part of this work and obtained some design guidelines for future game designers. We hope that our work will serve as the framework for the future explorations of making games better using 3D user interface technologies.

This thesis work has been part of an effort to improve the after-action reviews of the air combat simulator training sessions conducted at the Swedish Air Force Combat Simulation Centre (FLSC). Initial studies identified three main needs regarding the evaluation of air-to-air missile shots during beyond-visual-range combat simulation. These needs included an improved detection of where and when in the simulation playback a missile shot took place, a collected view of flight parameters to prevent confusion and cross-referencing between the various displays, as well as the ability to review an aircraft’s flight parameters over time in order to discuss alternative shooting opportunities or maneuvering patterns. To fulfill these three needs, design studies were performed iteratively in collaboration with staff at the FLSC. This work has resulted in a design proposal with a prototype based on the design guidelines and recommendations of the study's participants. The purpose of the visualization is to provide support for instructors and promote the individual learning of pilots. Hopefully, this can ultimately help in answering the question regarding why a missile missed its target. For instructors and air units such aids could mean that operating errors can be more easily identified and also form a basis for discussion during the assessment briefings. / Detta examensarbete har haft som syfte att förbättra utvärderingen av luftstridssimuleringar som bedrivs vid det svenska flygvapnets luftstridssimuleringscentrum, FLSC. Inledande studier identifierade tre huvudsakliga behov för utvärderingen av flygplansburna robotskott avfyrade mot luftmål utom synhåll, på långa avstånd. Dessa behov inkluderar en förbättring när det gäller att upptäcka var och när i en simuleringsuppspelning som ett robotskott har skett, en samlad vy över flygparametrar för att förhindra förvirring och korsreferering mellan olika skärmar, samt möjligheten att utvärdera ett flygplans flygparametrar över tid för att kunna diskutera alternativa avfyrningsmöjligheter eller manövreringsmönster. För att fylla dessa tre behov har iterativa designstudier utförts i samarbete med personalen på FLSC. Detta har resulterat i ett designförslag med en prototyp baserad på de designriktlinjer och -rekommendationer som studiens deltagare delgett. Syftet med visualiseringen är att ge stöd till instruktörer och främja piloters individuella inlärning. Förhoppningsvis kan detta i slutändan bidra till att svara på frågan om varför en robot missade sitt mål. För instruktörer och flygförband kan ett sådant hjälpmedel underlätta identifiering av felmanövreringar och även ligga till grund för värdefulla diskussioner under analysen av genomförda luftsstridsimuleringar.