An output-driven approach to design a swarming model for architectural indoor environments

•A novel swarming behavior model suitable for robust acquisition of architectural spaces.•An inverse modeling approach for determining swarming behavior model parameters. and•A practical design tool for predicting the suitability of a set of mobile sensors in a target environment. [Display omitted] Swarming Architectural Indoor Environments: Our output-driven approach consists of a novel swarming model component and an optimization component which uses a set of indicators to specify desired acquisition behavior. a) Visualization of the capture process of a swarm of 100 agents through a 750 m2 environment. Wall coloring indicates level of sampling (red=high) and shading indicates that the location has been visited. This global map is in fact not visible to the individual agents and it is only for visualization purposes. b) A depiction of the map of a single agent early on in the capture and c) is the map of the same agent near the end. During the acquisition, the agents do not actually share map information. We introduce a novel tool for designing a swarming behavior model for a set of virtual agents to automatically capture an initially unknown indoor architectural environment. Our key idea is to use an output-driven optimization to create targeted swarming behavior. The input to our model is a simple rectangular proxy of the target area and desired acquisition indicator values. The final outputs are the parameters for a swarming behavior model that is autonomous and decentralized, uses only local exploration, and is robust to agent failure. We show and compare the swarming performance in several simulated environments of up to several hundred square meters, 100 agents, and under various conditions.