COoperative Multiagent System for automated TArget Recognition by UAVs (COMSTAR)
Navy STTR FY2004


Sol No.: Navy STTR FY2004
Topic No.: N04-T005
Topic Title: COoperative Multiagent System for automated TArget Recognition by UAVs (COMSTAR)
Proposal No.: N045-005-0112
Firm: 21st Century Systems, Incorporated
12152 Windsor Hall Way
Herndon, Virginia 20170-2359
Contact: Plamen Petrov
Phone: (402) 505-7885
Web Site: www.21csi.com
Abstract: Unmanned Aerial Vehicles (UAVs) use has increased significantly, from rudimentary reconnaissance to complex missions. In the process of becoming smarter, UAVs have also become larger, more complex, and a lot more expensive. An alternative to the complex and costly UAV model is a new paradigm employing multiple, small or mini-UAVs, that operate in virtual swarms to achieve the complex objectives of today's reconnaissance missions. The advantage of this UAV swarm approach is clearly in the overall cost, increased redundancy, and a drastic reduction in single points of failure. The challenge is to develop swarming behaviors that can cope with the difficult conditions of combat. The team of 21st Century Systems, Inc. and the University of Nebraska, Omaha propose researching a concept called Co-Operative Multi-agent Automated Target Recognition UAVs (COMSTAR-UAV) to take advantage of advances in multi-agent system software ands the latest sensor and processing capabilities. The COMSTAR-UAV concept will be able to scale up efficiently in the number of UAVs deployed and in the amount of data collected and processed. Members of a COMSTAR-UAV swarm can congregate to exchange information and resources with each other making them more resilient to wireless network communication failures and loss of line of sight.
Benefits: The advantages of the swarmed-cooperative model in COMSTAR over the traditional method of uploading and processing data only at the central location are: (1) Most of the computation in the swarmed model is done in a distributed manner. Therefore, the system can scale up efficiently in the number of UAVs deployed and in the amount of data collected and processed for automated target recognition (ATR). (2) Since most of the computation in the swarmed model is done locally within the UAVs and not directly the central location, the functionality of the entire system is dependent to a lesser extent on the central location. This improves the dependability of the system in the swarmed model because a temporary failure (software or hardware) at the central location does not paralyze the entire system. (3) The swarmed model can perform ATR more rapidly by performing some computation locally on each UAV. This reduces the time required to upload every image data from different UAVs onto the central location and improves the efficiency of the system. (4) In a central location-based model every individual UAV communicate with the central location to upload data. Therefore, the system is susceptible to loss of data and failures due to wireless communication problems between an individual UAV and the central location that might be located far away. In contrast, UAVs in a swarmed model can congregate to exchange information and resources with each other without involving the central location unless necessary. Therefore, the swarmed model is more resilient to wireless network communication failures than the central location based model. (5) A swarm of dispersed UAVs can resolve an accurate position on a target more rapidly due to concurrent, distinct lines of position.

Return