Research Abstracts - 2007

Figures: (left) An example route from Bexley to Building 1; (right) An example visitor's agenda.

Introduction

The notion of situational awareness - where we are, where we are going, and how we are getting there - plays a significant role in our daily lives. Anyone who has ever been lost trying to negotiate an unfamiliar environment knows that the most familiar way to achieve situational awareness is by referencing a map of the local environment. Constructing such comprehensive models of local environments is central to the notion of giving both humans and robots alike a more keen awareness of their surroundings for navigational and analytic purposes, much like a street map is useful to travelers visiting an unfamiliar city.

Unfortunately, creating such models manually and regularly is infeasible due to the complexity and dynamics of everyday environments. This research project aims to automatically compile feature-rich, geometric, semantic and symbolic models of local environments organized from common data sources such as architectural floor plans. The models will then be integrated into human and robotic navigation systems, to the effect of greatly enhancing and demonstrating their usability. We propose to do this by: (1) developing a robust and scalable data representation that exhaustively characterizes geospatial features; (2) defining a common interface to access these data in a federated and asynchronous manner for simple and remote access by existing location-based systems; and (3) demonstrating the framework with a prototype application.

An existing demonstration of these powerful concepts is available in an application developed by our team entitled MIT Wikimap. This project is a collaborative effort to catalogue and share points of interest around the MIT Campus.

The prototype application to be developed is an agenda-based scheduling planner that allows hosts to create spatio-temporally constrained programs for visitors. The resulting schedules provide temporally- and spatially-optimized plans for visitors based on their intended agendas at the visited sites (Figure: right).

Approach

The development pipeline we describe above is known as the Building Model Generation (BMG) pipeline. The idea behind BMG is as follows: floor plans (in their native AutoCAD format) are inputted into the BMG pipeline, where the data are processed into a usable geometric format and catalogued into our representational framework. This geometric data is further analyzed for the purpose of deriving a topological layout of the environment in the form of a graph with nodes and edges. The geometry and topology is simultaneously tagged with semantic properties that give the data further richness in modeling the environment (for example, by denoting space types, portal types, and regions of semantically similar spaces). Combining these different spatial properties gives way to developing applications that can provide virtual tours of a real space or even automatically generate routes between two spaces (Figure: left) based on high-level queries.

References:

[1] V. Kulikov. Building Model Generation Project: Generating a Model of the MIT Campus Terrain. Master of Engineering Thesis, Massachusetts Institute of Technology, May 2004.

[2] P. Nichols. Location-Aware Active Signage. Master of Engineering Thesis, Massachusetts Institute of Technology, January 2004.