A Sensory Learning Scheme for Adaptable Mobile Agents

We propose a system architecture and methodology to manage the information flow and inquiries between a central station and an adaptable mobile agent (AMA) in real time. The agents collect and process sensory information from a target area in response to queries from the station and provide quantifiable responses to the station for evaluation and feedback. The station communicates with the AMA through Prolog implemented queries and commands, and the AMA performs detailed sensory data collection and processing. A key aspect of the proposed system methodology is the adaptation and learning capability which is embedded in the AMA. The AMA employs two onboard learning mechanisms. The first employs supervised query learning which is applied to the sequence of station queries, AMA evaluations and station feedback to the AMA. The second mechanism employs supervised and unsupervised training applied to onboard processed sensory data with the aim to provide adaptive recognition of target areas of interest. The AMA can collect more detailed data due to its ability to move into proximity with the target area. The proposed architecture addresses several defense and civilian oriented scenarios processing and evaluating query information from a central station to remote target areas in real time.