100 likes | 322 Views
Space Syntax & multi-agent simulation. An Exploration of Architectural Theory In Multi-Agent Simulation Glenn Elliott – Robotics 790, Fall 2008 . Overview. My project will investigate the applications Space Syntax, a theory from the field of Architecture, in multi-agent simulation.
E N D
Space Syntax &multi-agent simulation An Exploration of Architectural Theory In Multi-Agent Simulation Glenn Elliott – Robotics 790, Fall 2008
Overview • My project will investigate the applications Space Syntax, a theory from the field of Architecture, in multi-agent simulation. • The project will strive to replicate human movement patterns derived from the relation between humans and their environment’s spatial configuration.
Space Syntax • Space Syntax is an Architectural theory that claims space, the voids between objects (including walls, floors, etc.), can be described as a traditional graph of nodes and edges. • The major ramification of this is that computation may be performed on space-graph to show various spatial characteristics. • Some of these have a strong influence on how humans move through the described space.
Space Syntax Graphs Images from Space is the Machine by Bill Hillier. ^ Graphs describing the relations of blocks A and B to surface C in different configurations. Floor plans of the same area may have > radically different graphs.
Prior Art • The bulk of Space Syntax applications in multi-agent simulation has been done by Alan Penn and Alasdair Turner (see “Space Syntax Based Agent Simulation” in Proceedings of the 1st International Conference on Pedestrian and Evacuation Dynamics). • I have not yet found any other published contributors to Space Syntax based multi-agent simulations.
Prior Art – Penn & Turner’s Method • Step 1: Preprocess a 2D floor plan, deriving visibility of every point to every other point in the floor plan. This is known as a visibility map. (O(n2)) • Points are generated from a uniform sample across the floor plan. • Step 2: Store visibility information at each point in “angle buckets” such that queries on point visibility within a cone of vision can be quickly performed. • Example: “Give me all points visible at point X between -60 and 60 degrees of vision.” • Step 3: Place an agent in the environment with a cone of vision. Move the agent towards a random visible goal point within their cone of vision. Repeat every few steps.
Prior Art – Penn & Turner’s Method < An agent’s local view. Regions with high visibility have a high probability of being selected as a goal location. This effectively “draws” an agent into regions of higher visibility.
Prior Art – Penn & Turner’s Method < Frequency of many agents’ paths through an environment (Ikea store).
Novel Aspects • There appear to be promising areas of growth for Space Syntax in multi-agent simulation. • The “Visibility Map” is only one of many Space Syntax measurements. There are several other higher-level metrics that may be useful an agent simulation. • Penn and Turner approach multi-agent simulation from an Architect’s perspective. Their methods analyze an entire space/floor plan/building. • Requires a costly O(n2) pre-computation.
Goals • Add Space Syntax ideas into existing UNC RVO multi-agent simulator. • Evacuation scenarios. May enhance existing algorithms or develop on-the-fly methods. Agents may “explore” between waypoints towards building exit. • Hide-and-Seek. Requires higher level Space Syntax metrics (Axial Map, Depth Map). • Will not interfere with exiting RVO features such as Proxy Agents.