Persistent Surveillance of Transient Events with Unknown Statistics
Abstract
We consider the use of a mobile agent to monitor stochastic, transient events that occur in discrete locations in the environment with the objective of maximizing the number of event observations in a balanced manner. We assume that the events of interest at each station follow a stochastic process with an initially unknown and station-specific rate parameter. Consequently, the persistent monitoring problem we address in this paper is a bandit problem -similar to the canonical Multi-Armed Bandit problem- in which we are faced with the inherent trade-off between exploration and exploitation. We introduce a novel monitoring algorithm with provable guarantees that leverages variance estimates to generate policies capable of simultaneously taking into account the pertinent monitoring objectives and the balance between exploration and exploitation. We present analysis establishing lower bounds for the performance of our algorithm measured with respect to the quality of the policies generated. We present experimental results supporting our proposed algorithm and comparing its performance to that of current state-of-the-art monitoring algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 18, 2016
- Accession Number
- AD1033485
Entities
People
- Cenk Baykal
- Daniela L. Rus
- Guy Rosman
- Kyle Kotowick
- Mark Donahue
Organizations
- MIT Lincoln Laboratory