Computer Science Colloquia
April 25, 2012
Advisor: John Stankovic
Attending Faculty: Kamin Whitehouse, Chair; John Lach, Alfred Weaver, and Gabriel Robins
9:00 AM, Rice Hall, Rm. 504
Ph.D. Proposal Presentation
Decentralized Infrastructure-free Accurate Collaborative Localization for Firefighters
The firefighter's job is to rescue people and fight fires. During the rescue action, localizing each firefighter is critical, since the incident commander (IC) on the scene needs the location information to make tactical decisions, and firefighters need it for doing better rescue, cooperation and evacuation. After six firefighters died in 1999's Worcester Cold Storage and Warehouse Co. Fire because of being lost inside the building, the U.S. government has been pushing research institutes and companies to develop a localization system which can accurately and reliably track firefighters movements in the field.
The most advanced firefighter localization systems so far are the WPI Precision Personnel Locator (PPL) system and the Honeywell GLANSER system. The ideas of both systems are similar, using an inertial guidance system carried by each firefighter for short term navigation, and a centralized UWB-based ranging system for long term correction. Although they demonstrate the workability of their systems in small buildings, the drawbacks are: 1)their localization approaches require centralized communication and ranging schemes, however, in tall buildings or deep basements, it is reported that both the communication signal and the ranging signal are very likely to fail. 2) When firefighters are out of range of the base station, they independently rely on their own INS devices for localization without making use of other's position information, which causes the error to quickly accumulate. 3) The centralized architecture needs significant time for deployment and calibration, and the devices are expensive.
To solve these problems, we propose a novel inexpensive decentralized localization system for firefighters, which can be used stand alone or as a compensation for the centralized architecture. In this system, each firefighter carries a locator unit (LU) and a breadcrumb dispenser. The LU integrates an inertial measurement unit (IMU), a radio and a ranging device, while the dispenser's function is to dispense motes (breadcrumbs) on the ground, working as communication relays and landmarks. Each LU or breadcrumb is a networking node in our system. It only estimates its own position based on the local states, and sometimes it may exchange information with others when they meet (LU to LU, or LU to breadcrumb). The most important benefits of the system are: 1) it is fully decentralized and asynchronous, so it avoids the fragile centralized communication and ranging problem. 2) The system can achieve approximately best possible estimation through collaboration among LUs and breadcrumbs. 3) The static breadcrumbs play three roles in our system: landmarks for localization, relays for communication, and sensors for environmental monitoring. This not only improves the localization accuracy, but also increases system's functional and non-functional capability. 4) The system has very good scalability not only in terms of space, but also in terms of the number of users. This is because our efficient Elastic Decentralized Collaborative Localization (EDCL) algorithm is independent of the number of participants. 5) The system provides not only position estimates, but also the confidence information about these estimates, which could be used for localizing a firefighter in a certain area or selecting a retreat route based on breadcrumbs' location confidence.
Our system is divided into three parts: the EDCL decentralized fusion algorithm, the dead reckoning module and the inter-node ranging detector. The EDCL is an approximately optimal information-filter based estimator, whose performance is elastic to devices with different capabilities, and its resource consumption is independent of the total number of the participants. For the dead reckoning module, we directly use the shoe-mounted inertial navigation system (INS), since it is the most accurate and reliable INS so far for pedestrian tracking. For the inter-node ranging detector, we use RSSI to calculate the distance. Although it is well known that indoor RSSI is not accurate for distance mapping, we use three methods to reduce the multi-path effect, and make the final RSSI usable. The preliminary results, evaluation plan and the timeline are also presented in this proposal.