Mark Abramson

Estimating Social Network Structure and Propagation Dynamics for an Infectious Disease

The ability to learn network structure characteristics and disease dynamic parameters improves the predictive power of epidemic models, the understanding of disease propagation processes and the development of efficient curing and vaccination policies. This paper presents a parameter estimation method that learns network characteristics and disease dynamics from our estimated infection curve. We apply the method to data collected during the 2009 H1N1 epidemic and show that the best-fit model, among a family of graphs, admits a scale-free network. This finding implies that random vaccination alone will not efficiently halt the spread of influenza, and instead vaccination and contact-reduction programs should exploit the special network structure.

Asynchronous, distributed optimization for the coordinated planning of air and space assets

In this paper, we examine how to improve the frequency and accuracy with which highquality Earth observations are made by coordinating across multiple collection systems, including air and space assets, in an asynchronous environment. In particular, we consider how these improvements could impact Earth observing sensors in two use areas; climate studies and intelligence collection operations. To do this, we make simplifying but reasonable assumptions and use a complex yet intuitive value function to solve a series of simple optimization problems that allocate requests to single-mission planners, or “sub-planners.” We consider requests with time windows and priority levels, some of which require simultaneous observations by different sensors. The primary contributions of this paper include our approach to the asynchronous and distributed nature of the problem and the development of a value function to facilitate the coordination of the observations with multiple surveillance assets.

Coordination Manager - Antidote to the Stovepipe Antipattern

Many large data collection enterprises, (e.g., the collection of Earth observation data) are organized as single-mission systems of separately managed collection systems, i.e., stovepipes. This organizational approach inhibits communication and coordination among collection systems. The approach is so pervasive that essentially it is a pattern of behavior that new collection systems will be organized as a stovepipe. This pattern limits the collection potential of the overall enterprise – and so we view this pattern as a negative way of doing things or an anti-pattern. This paper discusses our solution to this anti-pattern: the coordination manager, which provides a framework for optimization-based rolling horizon dynamic planning and scheduling, an approach that addresses the issue of coordination of stovepipe mission planning systems. It is the first step toward an objective multiple mission system designed from the beginning to participate with other single-mission systems.

Robust Planning for the Earth Observing-1 (EO-1) Mission

We have developed a robust optimization-based technique for EO-1 image scheduling that uses Air Force Weather Agency (AFWA) Stochastic Cloud Forecast Model (SCFM) forecasts of cloud cover as an explicit input. This paper presents the details of the technique and demonstrates, using simulated operations based on actual EO-1 sites, historical ephemeris, and archived SCFM data, that using the technique significantly increases steady state value of the catalog of images.

Intelligent autonomy for small throwable land robots

DARPAs Tactical Mobile Robot program includes a Throwable Robot (Throwbot) designed to be thrown into buildings, then teleoperated for surveillance purposes. Use by ground troops imposes significant size and weight limits, as does the requirement that it survive ballistic delivery. The current program stresses the state of the art in robotics and packaging, but further challenges exist. Future Throwbots would benefit from significant increases in autonomy, to deal with RF communication difficulties in buildings and to allow simultaneous operation of multiple vehicles by one person. This paper describes both currently planned and advanced autonomous capabilities.

Earth Phenomena Observation System (EPOS) for Coordination of Asynchronous Sensor Webs

Draper’s Earth Phenomena Observation System (EPOS) has developed mission management solutions for sensor webs over the past decade. We are extending the EPOS capability to include coordination of asynchronous sensor webs, with a particular focus on enhancing both the quantity and value of the data available to scientists in developing science models related to hurricanes. We describe our efforts supporting the Hurricane Severe Storm Sentinel (HS3) mission. We discuss our approach to improving on key collection metrics of interest to hurricane scientists, e.g., persistent surveillance of measurements in the hurricane box, through greater use of available sensor web assets and optimization of coordination planning decisions, e.g., take-off time of HS3 mission aircraft.

Coordinated Planning of Asset Operations for CLARREO-like Missions

The recent development of more advanced sensor and communication systems has improved researchers’ abilities to monitor the Earth’s climate. However, current Earth observation operations are hindered by the existence of “stovepipe” systems in which inefficiencies are inherent. In this paper, we will apply a recently developed planning algorithm and run test cases to demonstrate the value of coordinating the observations/measurements of Earth-based targets across multiple collection systems. Specifically, we will compare coordinated and uncoordinated cases in a notional representation that uses a previously proposed version of the Climate Absolute Radiance and Refractivity Observatory (CLARREO) mission.

Robust Planning Under Uncertainty

** * * We consider the modeling and computational benefits of a recent approach to robust planning under uncertainty. For large problems, traditional stochastic decision formulations such as dynamic programming are often unsuitable because of computation time considerations. As a result, other reasonable approaches are deployed such as closed-loop planning and control, in which deterministic models and algorithms are invoked when there is a sufficiently disruptive change in the system data, e.g., in the environment. Our research examines an approach recently proposed by Bertsimas and Sim, which comes with theoretical performance guarantees. They develop a mathematical programming formulation that trades away some of the “optimal” performance in return for an increased probability that the solution will remain feasible despite disruptive changes in the system data. This formulation offers several advantages: few assumptions are made on the mathematical nature of the uncertainty, the formulation is applicable to a wide range of problem domains, and it has tractable computational complexity. We evaluate the Bertsimas-Sim formulation across a range of standard problems found in the literature and those that arise in a two-level UAV planning problem, observe several potential drawbacks of the current formulation, and propose modifications to address these issues.

Optimized Stochastic Coordinated Planning of Asynchronous Air and Space Assets

There are many organizations that use satellites and drones to collect information, such as ground photographs or atmospheric pressure measurements. Often, these separate organizations have overlap...

Optimized Coordinated Planning of Asynchronous Air and Space Assets in the Presence of Uncertainty

There are many organizations that use unmanned assets, such as satellites or drones, to collect information. This may include taking pictures of the ground, gathering infrared photos, taking atmospheric pressure measurements, or any conceivable form of data collection. Often these separate organizations have overlapping collection interests or flight plans that are sending sensors into similar regions. However, they tend to be controlled by separate planning systems which operate on asynchronous scheduling cycles. We present a method for coordinating various collection tasks between the planning systems in order to vastly increase the utility that can be gained from these assets. This method focuses on allocation of collection requests to scheduling systems rather than complete centralized planning over the entire system so that the current planning infrastructure can be maintained without changing any aspects of the schedulers. We expand on previous work in this area by inclusion of a learning method to capture information about the uncertainty pertaining to the completion of collection tasks, and subsequently utilize this information in a mathematical programming method for resource allocation. An analysis of results and improvements as compared to current operations is presented at the end.