You are here

Automated Generation/Learning of Discrete Event Simulation Models

Description:

OBJECTIVE: To develop machine learning and automated model generation methods to understand the flow of traffic (aircraft, people, and equipment) on carrier decks and related chaotic and constrained environments. The focus of this effort will be to develop methods that can utilize data from RFID tags, inexpensive cameras, and other inexpensive passive sensors to generate testable and scalable discrete event models of behavior from observation. Note that the focus of this effort is not on the development of new sensing hardware, sensor processing, or computer vision or tracking algorithms. Rather the solution should consider how the existing state-of-the-art in these areas could be used to generate data for the machine learning/model generation methods to model activity on the deck. DESCRIPTION: Aircraft handling aboard the flight and hangar decks of aircraft carriers is a series of complex processes in a constrained and chaotic environment. This includes recovery, refueling, payload loading, servicing and maintenance, manning, and positioning for the next launch. Understanding the flow of aircraft, people, and equipment through these processes is a challenge given the space constraints, available parking areas which change depending on the phase of ship operations, weather, operational factors, and constraints on people and equipment. As unmanned air systems begin to integrate in greater numbers it will be critical to understand this traffic flow better to allow for the development of wholly new paradigms that can optimize the use of both manned and unmanned systems, and support the increased tempo of operations and higher sortie generation rate requirements of future carriers. The focus of this effort will be to develop methods based on machine learning and automated model generation that can utilize data from RFID tags, inexpensive cameras, and other inexpensive passive sensors to generate testable and scalable discrete event models of behavior from observation. While work has been done with model generation in limited and relatively static domains, there are significant challenges in machine learning/automated model generation with large numbers of physical agents in an efficient and scalable manner and in such a complex environment. Advances have been made and demonstrated within simpler domains such as team sports with a limited number of players, collective animal behaviors, and logistics and inventory applications. While solving the general automated modeling problem with such a large number of moving entities would probably not yet be feasible, carrier operations have a structure that can be exploited to make this problem potentially feasible to solve. One significant challenge will be utilizing the type of data that is available from passive sensors. For example, interrogating passive RFID tags can be useful for identification and status information, but may have significant limitations in the precision of localization. As well, interrogation will be constrained by shipboard RF limitations. In contrast, vision systems may provide much greater localization precision and longer term tracking, but may have difficulties dealing with large and variable numbers of entities of interest, synchronizing multiple sensors, maintaining tracks in cluttered and occluded environments, and detecting relevant entities in the environment while avoiding non-relevant ones. Also, note that equipment refers only to mobile heavy equipment that impacts traffic flow on the decks (e.g., forklifts) and not tools and parts inventories. Tracking of smaller items that do not have significant impact on traffic flow is outside the scope of this topic. PHASE I: Propose a sensing method that would be appropriate for the environment of a carrier. Develop and implement an initial version of the proposed model generation/learning methods for a limited set of ship-like environmental factors with sufficient functionality to demonstrate feasibility. This could leverage data from simulations, pre-recorded data from activity in a constrained environment (with similarity to the carrier deck case with regards to particular characteristics), and/or very limited scope laboratory experiments with sensor hardware. Note that the use of actual data or data collection from an actual carrier is not required. A similarly constrained environment in a laboratory or outdoor environment with moving people and equipment would be sufficient. Develop metrics to evaluate the system in Phase II and determine an initial concept for how the approach could be used in a carrier environment. PHASE II: Further develop the proposed learning/model generation approach with a broader set of environmental conditions in a more dynamic and unstructured environment that has much of the complexity of a carrier environment. Then integrate these with a higher fidelity simulation and experiments with live assets in a laboratory environment. Refine the proposed sensing methods. Similar to Phase I, the use of data, or data collection, from an actual carrier is not required. A similarly constrained environment with moving people and equipment would be sufficient to test the feasibility of the approach. And, refine how the approach could be used in a carrier environment. PHASE III: Integrate the software with other components in a naval system and participate in integrated demonstrations of autonomous systems operations or in experiments with manned assets on a carrier deck. PRIVATE SECTOR COMMERCIAL POTENTIAL/DUAL-USE APPLICATIONS: This capability could be used in a broad range of military and civilian security applications of unmanned systems and in other applications involving management of automated systems, such as logistics, inventory, and industrial applications. REFERENCES: 1. Rochlin, Gene I., Todd R. La Porte, and Karlene H. Roberts. 1987."The Self-Designing High-Reliability Organization: Aircraft Carrier Flight Operations at Sea."Naval War College Review. http://govleaders.org/reliability.htm 2. Ryan, J. C. 2011."Investigating Possible Effects of UAVs on Aircraft Carrier Deck Crew Workload and Safety."Paper presented at the Human Systems Integration Symposium, Vienna, VA, Nov. 2011. http://web.mit.edu/aeroastro/labs/halab/papers/JRyanHSIS-2011.pdf 3. Ryan, J.C., M.L. Cummings, Nick Roy, Ashis Banerjee and Axel Schulte. 2011."Designing an Interactive Local and Global Decision Support System for Aircraft Carrier Deck Scheduling."Paper presented at the AIAA Infotech@Aerospace 2011, St. Louis, MO, March 2011. http://www.mit.edu/~ashis/Articles/Ryan_AIAAInfotech_2011.pdf 4. Dollar, Piotr, Christian Wojek, Bernt Schiele and Pietro Perona."Pedestrian Detection: An Evaluation of the State of the Art."Paper submitted to IEEE Transactions on Pattern Analysis and Machine Intelligence. http://vision.ucsd.edu/~pdollar/files/papers/DollarPAMI11peds.pdf 5. Chellappa, Rama, Amit K. Roy-Chowdhury and S. Kevin Zhou. 2005. Recognition of Humans and Their Activities Using Video. Synthesis Lectures in Image, Video, & Multimedia Processing, Morgan & Claypool Publishing. 6. Feldman, Adam, Maria Hybinette, Tucker Balch and Rick Cavallaro. 2012."The Multi-ICP Tracker: An Online Algorithm for Tracking Multiple Interacting Targets."Journal of Field Robotics. http://www.cc.gatech.edu/fac/Tucker.Balch/Papers/2011-multi-icp.pdf 7. Guillory, Andrew, Hia Nguyen, Tucker Balch and Charles L. Isbell, Jr. 2006."Learning Executable Agent Behaviors from Observation."DOI: 10.1145/1160633.1160774 8. Bio-tracking: Tracking and understanding animal behavior. http://www.bio-tracking.org/ 9. Balch, Tucker, Frank Dellaert, Adam Feldman, Andrew Guillory, Charles Isbell, Jr., Zia Khan, Stephen C. Pratt, Andrew N. Stein and Hank Wilde. 2006."How Multirobot Systems Research Will Accelerate Our Understanding of Social Animal Behavior". Vol 94(7) pp. 1445--1463, 2006. http://www.cs.washington.edu/homes/guillory/ieee06.pdf
US Flag An Official Website of the United States Government