We envision a future in which conversational virtual agents collaborate with humans in games and training simulations. A representation of common ground for everyday scenarios is essential for these agents if they are to be effective collaborators and communicators. Effective collaborators can infer a partner’s goals and predict future actions. Effective communicators can infer the meaning of utterances based on semantic context. This article introduces a computational model of common ground called a Plan Network, a statistical model that encodes context-sensitive expected patterns of behavior and language, with dependencies on social roles and object affordances. We describe a methodology for unsupervised learning of a Plan Network using a multiplayer video game, visualization of this network, and evaluation of the learned model with respect to human judgment of typical behavior. Specifically, we describe learning the Restaurant Plan Network from data collected from over 5,000 gameplay sessions of a minimal investment multiplayer online (MIMO) role-playing game called The Restaurant Game. Our results demonstrate a kind of social common sense for virtual agents, and have implications for automatic authoring of content in the future.
In this article, we address the particular issue of authoring interactive narrative with respect to video-games and interactive storytelling . We first introduce the narrative paradox between interactivity and narrative content in virtual environments and consider its impact on game design and development. We then introduce the concept of the Emergent Narrative (EN) and the particular phil osophy it has been developed upon. Finally, we describe an authoring process for this approach that reflects on the characteristics of interacting within such a narrat ive framework.
3D video games use texture maps to improve the realism and the visual detail of graphical objects without significantly increasing rendering complexity. The general trend in the gaming industry towards the use of large texture maps has lead to the use of texture compression techniques. Current color texture compression techniques treat all texture content uniformly which can lead to reduced visual quality. This paper addresses this issue by introducing a perceptually adaptive approach to color texture map compression that is based on the human vision system. The proposed method attempts to improve visual detail and compression performance without adding computational complexity. Experimental results show improved visual quality and compression performance over existing techniques. The proposed method is well suited for real-time applications such as 3D video games.
Video game development is used to teach collaborative software engineering principles. However, when the collaboration is exclusively between computer scientists, a balkanized perspective is unintentionally promoted. A multidisciplinary faculty addressed how to exploit video game development as a vehicle for a collaborative cross-disciplinary experience in technology development for upper-level students from our contributing majors. This paper addresses the issues of curriculum structure, student assessment and cross-disciplinary team teaching. We present a dual model of cross-disciplinary content and individualized deliverables that allows each student to determine how narrowly or broadly to focus his/her personal learning.
This paper addresses an initial game design process focused on user input methods. This process led to experimental bimanual mouse gameplay.
Interactively rendering geo-spatial scenery is a recurring theme in game development. This paper demonstrates a method of handling much of that rendering on programmable graphics hardware. This makes better use of the power available on the GPU, and frees the CPU processing time for user interaction, simulation, and data acquisition and manipulation. Introduction and Previous Work There have been many methods proposed to rendering large terrains in real-time computer graphics systems. The amount of data available to render in terrains has always exceeded the available rendering time, making the issue of terrain level-of-detail an old problem. Most algorithms for terrain rendering focus on optimizing mesh-based algorithms for removing or inserting vertices based on some error criteria. A common method of handling terrain LOD is to build a large mesh, and selectively pick a vertex to add or remove from the mesh. When using a View Independent Progressive Mesh (VIPM) [6], the order of vertices to remove or add is precomputed, and does not depend on the camera position. For large terrains, where large areas may not be visible at all, VIPMs are less applicable. VIPMs are more typically applied to non-terrain meshes. View Dependent Progressive Meshes (VDPMs), as the name suggests, determine at run-time which vertices to remove or add. Typically, vertices are selected by determine the screen-space error induced by removing or adding the point. There have been many papers on VIPMs and VDPMS, such as [4], [7], [9]. VIPMs and VDPMs continually alter the rendered geometry, so the mesh must be retransmitted to the GPU whenever the detail needs to change. Transmitting this data takes a noticeable amount of time, and so we prefer to use static geometry, stored in Vertex Buffer Objects (VBOs) as often as possible. When using VBOs, OpenGL can move the geometry to the GPU’s memory, and avoid the transfer time from the CPU. A goal of this project was to ensure that our scheme can utilize the massive performance boost gained by using static VBOs. Rather than operate at the level of a single vertex, geomipmapping [3] breaks the large terrain into square chunks, also called zones. Various levels of detail are pre-generated for each zone, independent of the view. At run-time, a simple error metric is calculated for each zone to determine which LOD to render. Typically, this metric is based solely on the distance from the eye position to the center of the zone. As different zones may be operating at different levels of detail, care must be taken to ensure that there are no T-intersections between zones, which can cause cracks to appear between zones. Unlike progressive meshes, each LOD can be stored in a static VBO, as it never changes. This method uses a large quantity of GPU memory, so very large terrains require more elaborate managerial schemes. LODs for zones can be paged in at appropriate times, so that only the required level of detail is present. The zones can be maintained in a quad-tree, rather than a regular grid, which greatly aids GPU storage of larger terrains. Geomipmapping uses a very simple decimation of the geometry. The highest LOD contains every vertex, while every successively lower LOD skips a vertex. Each LOD, then, has roughly four times fewer points than the level higher. The name is an allusion to texture mip-mapping [5] which is appropriate, considering the similarity of the technique. Like progressive meshes, there have been other papers on this type of LOD, such as [22]. 1 A T-intersection occurs when a vertex is placed along an edge of an adjacent triangle, as opposed to being placed as a vertex of both triangles. This can cause rendering artifacts due to inaccuracies in the rasterizer. In the case of heightfields, large cracks will appear where T-intersections are allowed.
This paper constitutes a sequel to our previous work focused on investigating cooperative behaviors, adaptive learning and on-line interaction towards the generation of entertainment in computer games. A human-verified metric of interest (i.e. player entertainment) of predator/prey games and a neuroevolution on-line learning (i.e. during play) approach are used to serve this purpose. Experiments presented here demonstrate the generality of the proposed approach, in its ability to overcome the difficulties of learning in real-time and generate interesting predator/prey games, over game type, complexity, topology, initial opponent behavior and player type.
Lighting assumes many aesthetic and communicative functions in game environments that affect attention, immersion, visibility, and emotions. Game environments are dynamic and highly unpredictable; lighting such experiences to achieve desired visual goals is a very challenging problem. Current lighting methods rely on static manual techniques, which require designers to anticipate and account for all possible situations and user actions. Alternatively, we have developed ELE (Expressive Lighting Engine) -- an intelligent lighting system that automatically sets and adjusts scene lighting in real-time to achieve desired aesthetic and communicative goals. In this paper, we discuss ELE and its utility in dynamically manipulating the lighting in a scene to direct attention, stimulate tension, and maintain visual continuity. ELE has been integrated within Unreal Tournament 2003 . The videos shown at [14] shows a demonstration of a first person shooter game developed using the Unreal 2.0 engine, where ELE was configured to dynamically stimulate tension, while maintaining other visual goals.
