Computer generation of pictorial explanations

Steve Feiner · 1987

Although much work has been done on planning and generating textual explanations, relatively little attention has been devoted to automating the design of pictures that could be used to explain things. This thesis presents a three part conceptual architecture for generating pictorial explanations, such as maintenance and repair manuals, in which pictures play a crucial role. One part determines the material to be presented, for example, the solution to a maintenance and repair task. A second part plans the explanation, determines the spatial and temporal layout of the displays, and designs the pictures and text. A third part interfaces to rendering and typesetting software that scan-converts the displays. APEX (Automated Pictorial EXplanations) is a partial testbed implementation of the architecture. It can generate sequences of pictures that show how to perform simple actions, such as turning or pulling, in a world of 3D objects. Each picture crystallizes around those objects that participate directly in the action being depicted. Whether additional objects will be included and how they will be rendered are determined by the relationships that the objects bear to those already in the picture. For example, an object that is similar to one of the participating objects is added with enough detail to disambiguate it and is drawn with a rendering style intended to mark it as less important. Although APEX needs to depict objects at varying levels of detail, the original object database is a hierarchy with physical objects located only at the leaves. Selective display of leaves, however, will not produce effective pictures. A method for controlling the level of detail in a picture is presented in which the object hierarchy is processed to yield one whose internal nodes are associated with automatically created approximations of their children. Decisions made during the picture design process control the depth to which different parts of this hierarchy are traversed, determining the level of detail with which each part is rendered.

Read the paper · More papers on PaperTik