Multi-Agent Sensor Data Collection with Attrition Risk. Hudack, J. & Oh, J. In
Multi-Agent Sensor Data Collection with Attrition Risk [link]Paper  abstract   bibtex   1 download  
We introduce a multi-agent route planning problem for collecting sensor data in hostile or dangerous environments when communication is unavailable. Solutions must consider the risk of losing agents as they travel through the environment, maximizing the expected value of a plan. This requires plans that balance the number of agents used with the the risk of losing them and the data they have collected so far. While there are existing approaches that mitigate risk during task assignment, they do not explicitly account for the loss of agents as part of the planning process. We analyze the unique properties of the problem and provide a hierarchical agglomerative clustering algorithm that finds high value solutions with low computational overhead. We show that our solution is highly scalable, exhibiting performance gains on large problem instances with thousands of tasks.
@inproceedings {icaps16-52,
    track    = {​Main Track},
    title    = {Multi-Agent Sensor Data Collection with Attrition Risk},
    url      = {http://www.aaai.org/ocs/index.php/ICAPS/ICAPS16/paper/view/12994},
    author   = {Jeffrey Hudack and  Jae Oh},
    abstract = {We introduce a multi-agent route planning problem for collecting sensor data in hostile or dangerous environments when communication is unavailable. Solutions must consider the risk of losing agents as they travel through the environment, maximizing the expected value of a plan. This requires plans that balance the number of agents used with the the risk of losing them and the data they have collected so far. While there are existing approaches that mitigate risk during task assignment, they do not explicitly account for the loss of agents as part of the planning process. We analyze the unique properties of the problem and provide a hierarchical agglomerative clustering algorithm that finds high value solutions with low computational overhead. We show that our solution is highly scalable, exhibiting performance gains on large problem instances with thousands of tasks.},
    keywords = {Adversarial planning,Distributed and multi-agent planning,Planning activities; motions and paths}
}

Downloads: 1