Universal self-learning tool for developers VR applications, enabling automated and easy implementation of multiplayer capabilities in VR games on various devices.Asset Motion Recognizer requires python 3 installed. More information can be found in the Samples/MVN_MotionRecognition/README.md fileThe aim of the project is to develop a new product in the form of a universal self-learning tool for VR application developers enabling automated and easy implementation of multiplayer gameplay options in VR games on various devices using advanced artificial intelligence and machine learning algorithms that learn the specifics of a given programmer's work in order to propose the best solutions for a given project specificity.The developed artificial intelligence algorithms learn the specifics of a given programmer's work in real time designing a multiplayer VR game and generate dynamic hints and recommendations in the selection of the most effective methods and paths of the programmer's conduct in order to save time for implementation, and suggesting the implementation of functions most adequate for a given specificity project. The tool will allow for a significant acceleration of the programmer's work and will facilitate the organization of the source code. The use of machine learning techniques will also increase the realism of mapping the behavior of all characters participating in a multi-person VR experience through the ability to register and automatically create a library of movements based on recorded real player/user movements.The developed solution will be addressed to development teams and individual creators dealing with programming multiplayer gameplay in VR games and applications. Creation solutions currently available on the market Multiplayer games are not suited to creating VR experiences. Therefore, the creators use them in limited scope, and many functionalities have to be created independently from scratch. The solution proposed in the application It will therefore meet the direct needs of programmers designing multiplayer games in the VR environment.---The package contains additional examples added as unitypackage to better organize the file structure. The instructions in the main README file describe how to unpack the given packages.Features:Creating your own organizations and applicationsManage your applicationsCreate your own rooms, manage rooms and custom propertiesSupported OS: The application was tested in an editor on Windows. Builds were tested on Windows, Android and WebGL operating systems. Tested using Meta Quest 2, 3, HTC Vive gogglesDocumentation: https://vagency.smarthost.pl/mvn-doc/A broader description of the use of artificial intelligence in assets is included in the following scientific articles:MVN: an AI-powered multiplayer networking solution for VR gamesSupporting Unity Developers with an AI-Powered Asset: Insights from an Exploratory User Study on Multiplayer Game DevelopmentDesign and implementation of automation - development of an artificial intelligence engine based on training data collected as part of the asset creation, and then validation in a laboratory environment based on the test part of the data (so-called cross-validations). The task will include the selection of architecture, development of the knowledge base structure and development of a proprietary artificial intelligence (SI) algorithm supporting implementation tasks, offering the following scope of functionality:• Detection and suggestions for correcting common programming errors• Suggesting ways to implement solutions related to network communication, multiplayer and VR• Suggesting solutions that ensure scaling, e.g. different solutions for "intimate" online games and others (e.g. distributed) for a larger number of users.Building the knowledge base will be based on:• Training the AI engine with the help of existing code containing errors and its corrected and verified version;• Universal solutions proposed by various programmers regarding network and VR issues;• taking into account the so-called "good programming techniques" (keeping the program code transparent, using clear variable names, avoiding code repetition, etc.);• The AI engine should take into account and propose the use of design patterns;• The suggested solutions should be consistent with the agile software development methodology and take into account rules allowing for the generation of simple, modular and clear program code (KISS, DRY, YAGNI, TDA or SOC principles, etc.).---Development of an artificial intelligence engine based on collected test data. In order to identify player behavior patterns, a method will be developed to discover sequence patterns in recorded player behavior. For this purpose, a method for assessing sequence similarity will be developed sequence search method. Recognized behavior patterns will be saved in the database. Then a machine learning mechanism will be developed to enable prediction of the most probable one traffic. As part of the task, proprietary algorithms were developed:player behavior segmentation algorithm: this mechanism will allow mapping of descriptive time sequences player's move into a sequence of individual moves. A single move can be a sequence of changes between the same one player position or between static player positions.movement pattern discovery algorithm: the algorithm will be based on methods for discovering sequence patterns known from data mining methodologies. However, it should be taken into account that the sequences will be descriptive time sequences subsequent player positions, and the need to interpolate movement must be taken into account. For this purpose it is envisaged the use of Big Data methods (clustering) as an unsupervised learning method. This way the system itself will learn movement patterns as patterns defined by discovered cluster points.movement prediction algorithm: based on the analysis of the sequence of movements, the system will acquire knowledge about the most probable movements. It is planned to use the reinforcement learning method here learning). The system will be initially trained with test data, and then, during the programmer's work, with rules decisionmaking will be adjusted according to the programmer's decisions. In terms of initial learning, planning is done using Big Data methods to discover correlations between the order of individual movements or discovering probable sequences of movements. Knowledge will be stored in the system in the form of trees decisionmaking or a set of rules. During the operation of the system, prediction of subsequent movements will be made, the remembered rules will be modified depending on the consistency of the decisions made by the programmer.
