Asymptotic Analysis for Greedy Initialization of Threshold-Based Distributed Optimization of Persistent Monitoring on Graphs

Event-Driven Receding Horizon Control For Distributed Persistent Monitoring in Network Systems

“…., is a cyclic order of three elements: {−(B i − A i ), 0, A i }. As pointed out in [6], allowing overlapping dwell sessions at some target (also known as "target sharing") is known to lead to poor performance (clearly, this concern does not apply to single-agent problems). This observation motivates the next assumption.…”

“…To make sure that each agent is capable of enforcing the event [R i → 0 + ] at any i ∈ T , the following simple stability condition is assumed (similar to [6]).…”

“…Since we will adopt a graph topology for this problem, the condition s a (t) − Y i ≤ r a is represented by the agent residing at the ith vertex of a graph and and N i (t) is used to replace the role of the joint detection probability of a target i used in [7], [14]. Moreover, similar to [6] the analysis in this paper is independent of the agent motion dynamic model. c) Objective: Our objective is to minimize the mean system uncertainty J T over a finite time interval t ∈ [0, T ]:…”

“…A persistent monitoring problem arises when a dynamically changing environment is monitored by a set of mobile agents and encompasses applications such as environmental sensing [1], surveillance systems [2], energy management [3] and data collection [4]. In contrast to cases where every point in the environment is equally valued for agents to monitor, in many others, only a finite set of "points of interest" (henceforth called "targets") holds a positive value [5], [6]. The persistent monitoring problem considered in this paper belongs to the latter class, where the goal of the agent team is to monitor (sense or collect information from) each target in order to reduce an "uncertainty metric" associated with the target state.…”

“…However, this IPA-TCP approach often converges to poor locally optimal solutions. As a remedy, [6] has proposed to append an off-line centralized threshold initialization scheme which is shown to considerably increase performance at the expense of significant computational effort. Moreover, since on-line control is still governed by the IPA-TCP method, any state perturbation would trigger a new parameter tuning process with a considerable amount of recovery time.…”