Task 3.2: Localization in TerraSwarm Systems

[Dutta, Kumar, Pappas, Rowe]

Knowledge of the location of mobile devices will be important for many TerraSwarm activities. Many TerraSwarm components (assets) are dynamic in the sense that that they appear, disappear, and physically move. For example, a TerraSwarm application may include mobile sensors for collecting information, mobile vehicles for transport of physical objects and/or people, and mobile resources for network communication. Each asset must be capable of operating in isolation, in an ad-hoc group formed with neighboring assets, and with the cloud. It must be capable of forming "beliefs" about the environment in which it operates, leveraging any available (possibly sporadic) communication to reinforce and refine these beliefs by aggregating data from multiple sources.

Unfortunately, existing location tracking services based on GPS and WiFi tend to perform poorly indoors, especially when trying to define semantic (logical as opposed to physical) locations within a space. For example, if a person is standing close to a wall in a room, normal quantitative errors could result in placing the person in the wrong room (or even in mid-air, outside the building).

The focus of this task is on extraction of semantic location information from quantitive location information, and on development and evaluation of technologies that are better at semantic localization. Specifically, semantic localization is the problem of determining where in a logical space a person or thing is located, rather than where in physical space. The approach will combine solutions involving measurement technologies and machine learning techniques (e.g., solutions that fuse semantic information about buildings and behavioral patterns with location information). Technologies like VLC, UWB, BLE and ultrasound have proved to be viable indoor ranging candidates. For these systems to be adopted at scale, we need automated and robust approaches for configuration, map generation and annotation of semantic localization services. This will likely involve crowd-sourcing from mobile devices with input from indoor robotic systems.