G. Michael Youngblood

MavHome: an agent-based smart home

The goal of the MavHome (Managing An Intelligent Versatile Home) project is to create a home that acts as an intelligent agent. In this paper we introduce the MavHome architecture. The role of prediction algorithms within the architecture is discussed, and a meta-predictor is presented which combines the strengths of multiple approaches to inhabitant action prediction. We demonstrate the effectiveness of these algorithms on smart home data.

A multi-agent approach to controlling a smart environment

The goal of the MavHome (Managing An Intelligent Versa- tile Home) project is to create a home that acts as a rational agent. The agent seeks to maximize inhabitant comfort and minimize operation cost. In order to achieve these goals, the agent must be able to predict the mobility patterns and device usages of the inhabitants. Because of the size of the problem, controlling a smart environment can be effectively approached as a multi-agent task. Individual agents can address a portion of the problem but must coordinate their actions to accomplish the overall goals of the system. In this chapter, we discuss the application of multi-agent systems to the challenge of controlling a smart environment and describe its implementation in the MavHome project.

Seamlessly engineering a smart environment

Developing technologies and systems for automated control of home and workplace environments is a challenging problem. We present a complete agent architecture for learning to automate a smart environment and discuss integration of AT and middleware technologies necessary to achieve the goals of this project. Results are demonstrated using the MavPad and MavLab intelligent environments.

Influence points for tactical information in navigation meshes

Good artificial intelligence for strategy and first person shooter games requires tactical information. Tactical information assists agents in choosing appropriate places to place vulnerable resources, safe paths for moving through the world, and the most effective places to attack the enemy. Unfortunately, providing this information can be very expensive, as it typically requires keeping one or more full map arrays around. In this paper, we present influence points, a technique for providing tactical information by annotating navigation meshes. This greatly improves the memory cost of storing the information, and reduces the cost of updates as the world changes.

Understanding information observation in interactive 3D environments

Communicating information to the user is a vital part of the interactive experience. In order to better convey the information to the end user, we must know where to place this information and how to present it in a manner that it will be noticed. Subjectively placing this information is not sufficient since every user will interact with the environment in their own unique manner. Information value is a metric that provides us with the knowledge of which surfaces players looked at most in the environment in the form of an ordered list of surfaces. Using an empirical algorithm for discovering the information value of environmental surfaces from recorded player data, we performed a 150 subject information value study and found that placing information in the high value surfaces yields up to 60% improvement in user observation. However, most players did not recall the information that they had seen. We conducted another 150 subject study to investigate what factors improve information retention and found that popular images do improve recall by up to 28%. Finally, we conducted a 30 person study on the effect of changing the players task (context) from search to exploration on information recall and found that recall increased by 38%.

Representational complexity of reactive agents

Reactive agents are an important part of video games and numerous tools have emerged to facilitate the rapid construction of such agents. While the ability of the commonly used reactive techniques to express agent specifications is roughly equivalent, the authorial burden of constructing these specifications varies. In practice, this means that identical agent behavior may be more difficult to create in some architectures than others. In this paper we introduce the notion of representational complexity that relates to the authorial burden of constructing such agents and theoretically compare the representational complexity of finite state machines, behavior trees, and subsumption architectures. Our key finding is that hierarchical subsumption architectures have significantly lower representational complexity as compared to hierarchical finite state machines and behavior trees, which makes subsumption the best choice when developing authoring tools for non-expert users.

Efficacy of a Smartphone System to Support Groups in Behavior Change Programs

Smartphone platforms provide an excellent opportunity for projecting existing or new behavior-change methods into everyday life at great economies of scale. In this paper we present an experimental test of a new behavior-change smartphone platform and application called Fittle, which delivers ecological momentary interventions and group support to help people progressively master healthy habits. An 8-week field study involving 19 participants demonstrated the engagement and efficacy of Fittle across three classes of behavior (diet, physical activity, and stress-reduction). Individual adherence to the behavior programs was found to be associated with group membership. Content analysis of intragroup interactions suggests that high performance groups were generally more social, more supporting of each other on program goals, and shared more.

Optimal information placement in an interactive 3D environment

The correct placement of important artifacts and information in interactive three-dimensional (3D) environments is important to ensure that those key artifacts and information are seen and absorbed by the immersed user. This can include training information, advertisements, clues, interaction points, and other information that needs to be conveyed to or manipulated by the user. We propose a novel algorithm for calculating the optimal positioning of such artifacts and information based on a corpus of prior play testers, which are used to determine distance-weighted and radially focused observation densities on surfaces of interactive 3D environments. We have developed a tool called HIIVVE (Highly Interactive Information Value Visualization and Evaluation) which allows for interactive evaluation as well as offline processing of the information value surfaces. A user study involving information placement using the calculated information value surfaces and observation densities shows that higher valued locations do yield improved user observation by as much as 58.3%.

Spatial cues in hamlet

Many games, films, and virtual environments are very scripted, or use very canned/explicit cut scenes for characters to interact. This requires extensive work for producing new scenes, actions, and other scripts for these environments. It also usually comes with a certain level of expertise in lower-level character control. Current research focuses primarily on the conversational and non-verbal domains of this issue. However, with the growing focus on realistic virtual environments, the spatial domain is becoming a more critical component in creating that realism. Tools and markup languages, such as Behaviour Markup Language (BML), Functional Markup Language (FML), and BML Realizers are making it possible to abstract the control of virtual characters to a certain extent. Unfortunately, these methods still require a level of expertise and time that can be unreasonable. Therefore, we propose a higher level of abstraction to ease this authorial burden for new scenes and actions in games. To do this, we look to another example of scripted activities that has been used for hundreds of years: play-scripts. We took the fully annotated play-script for Gielguds Hamlet in 1964, along with the recording of the same play to create a baseline using BML. We compared this to the use of just the play-script and some very simplistic natural language processing to block the same act of the same play. The results of this comparison show a savings of over four hours for authoring spatial scripts, while maintaining similar spatial blocking.

Embedding Information into Game Worlds to Improve Interactive Intelligence

Current game worlds are visually rich but information poor – particularly poor from the artificial intelligence (AI) point of view. Where the player sees a rich visual representation of 3D objects, internally these are just very sparsely described collections of points in space. Tools for advanced world creation, character modeling, animation, and advancements in computer graphics have brought us into the age of near photo-realistic interaction; however, these interactions are still very limited in comparison to the real world, and the information is presented overwhelmingly for the player, packaged for the Graphics Processing Unit (GPU) with little reflection or structure suitable for use by AI systems. This problem of a lack of rich information suitable for consumption by the game AI often limits the true potential for deeper levels of interaction that are becoming more in-demand by game players. This chapter presents a number of tools and techniques, which are being used to improve the embedded information contained in immersive game worlds. Symbolic annotation of the environmental elements, advanced spatial decomposition, calculating the information value of the surfaces in an interactive environment, and visual analysis form the core tools and information generators of our Common Games Understanding and Learning (CGUL) Toolkit. Using these tools to incorporate information into the game design and development process can help create information-rich interactive worlds. AI developers can work with these environmental information elements to improve non-player character (NPC) interactions both with the player and the environment, enhancing interaction, and leading to new possibilities such as meaningful in-game learning and character portability. Case studies from two different projects using these techniques provide some additional insight and reference as to how these techniques have been incorporated into current game AI and research.