Alvaro Enrique Gil

Cooperative scheduling of tasks for networked uninhabited autonomous vehicles

In this paper we model a cooperative control problem for a network of UAVs where it is assumed that before the mission starts a set of tasks is given to a set of UAVs, but then after deployment the UAVs must cooperate to decide which UAV should process each task. The cooperation must occur in spite of imperfect communications (e.g., messages with random but bounded delays). We show how to view this as a cooperative scheduling (resource allocation) problem, and how to derive bounds on mission-level performance metrics for cooperative scheduling methods. Simulations will be used to compare the approach to a noncooperative strategy and to provide design guidelines for the cooperative scheduler.

Experiments for dynamic resource allocation, scheduling, and control: new challenges from information technology-enabled feedback control

To help educators design laboratories that support the study of information technology-enabled distributed feedback control, several inexpensive experiments for studying the design and implementation of distributed and networked dynamic resource allocation, scheduling, and control strategies are described. In each case, an overview of the experimental apparatus and the challenges involved were presented. For three of the testbeds, the experimental results for distributed decision-making strategies were shown. The ideas for experiment redesign highlight the existence of tradeoffs among experimental apparatus design, the choice of hardware and software, and research and educational objectives.

Stable Cooperative Surveillance With Information Flow Constraints

We consider a cooperative surveillance problem for a group of autonomous air vehicles (AAVs) that periodically receives information on suspected locations of targets from a satellite and then must cooperate to decide which AAV should search for each target. This cooperation must be performed in spite of imperfect intervehicle communications (e.g., messages with random but bounded delays), less than full communication connectivity between vehicles, uncertainty in target locations, and imperfect vehicle search sensors. We represent the state of the search progress with a ldquosearch map,rdquo and use an invariant set to model the set of states where there is no useful information on target locations. Arrivals of new suspected target location information from the satellite corresponds to perturbations of the search map from this invariant set. A cooperation strategy that pursues a type of ldquopersistent area denialrdquo will try to force trajectories of the system into the invariant set by exploiting initial target information and search progress by the AAVs. We show that the invariant set is exponentially stable for a class of cooperative surveillance strategies. We provide a comparative analysis of cooperative and noncooperative strategies. Next, we show via simulations the impact of imperfect communications, imperfect vehicle search sensors, uncertainty in search locations, and pop-up suspected locations on performance.

Blind Source Separation by Nuclear Norm Minimization and Local Recoverability Analysis

We propose a new blind source separation (BSS) algorithm that is effective when Hankel matrices constructed from individual source signals are near low-rank and satisfy a certain near-orthogonality condition. Source separation is achieved by finding a nonsingular reverse-mixing operation that minimizes nuclear norms of Hankel matrices constructed from estimated source signals. The new formulation results in a non-convex optimization problem involving a reverse-mixing matrix. Preliminary analysis of local recoverability of source signals as well as few numerical simulations are presented in this letter.

An agent-based model for crowdsourcing systems

Crowdsourcing is a complex system composed of many interactive distributed agents whom we have little information about. Agent-based modeling (ABM) is a natural way to study complex systems since they share common properties, such as the global behavior emerging on the basis of local interactions between elements. Although significant attention has been given to dynamics of crowdsourcing systems, relatively little is known about how workers react to varying configurations of tasks. In addition, existing ABMs for crowdsourcing systems are theoretical, and not based on data from real crowdsourcing platforms. The focus of this paper is on capturing the relationships among properties of tasks, characteristics of workers, and performance metrics via an ABM. This approach is validated by running experiments on Amazon Mechanical Turk (AMT).