ALIVE - Artificial Life in Virtual Environments
Nearly 50 years ago, Sydney Brenner introduced Caenorhabditis elegans as a model for studying developmental biology and neurology. Because of its simplicity, it has become one of the best understood organisms on the planet being the only one to have its cell lineage, genome, and nervous system completely mapped. Yet despite all of the effort that has gone into uncovering the secrets behind "the mind of the worm," we still lack a compelling systems-level model for how the neurons and the connections between them generate the surprisingly complex range of behaviors that are observed in this relatively simple organism.
The ALIVE project seeks to rectify this deficiency by developing the first-ever complete model of the body and nervous system of C. elegans in a virtual environment that mirrors the physical properties of its natural world.
Unmanned Aerial Vehicle Planning
Robert Burns once wrote: “The best-laid plans of mice and men oft go astray.” Nowhere is this statement truer than when planning missions in an uncertain, adversarial environment. Despite our best efforts and hundreds of years of experience, planning is still an arduous task whose results often dictate the outcome before the first action is ever taken. However, as Burns points out, even the best laid plans must be able to adapt to unforeseen circumstances to ensure that they succeed. The overall goal of this project is to develop new technologies to dynamically control and coordinate multiple autonomous vehicles so they can accomplish their missions while the enemy is attempting to deny them access and prevent them from communicating.
Our approach to addressing this complex problem is to expand upon our previous task allocation results by augmenting dynamic, distributed constraint reasoning with machine learning techniques and adaptive response strategies. By combining these technologies, we propose to build a system that can 1) develop robust, adaptable mission plans 2) exploit knowledge learned through prior interactions with our adversary, and 3) autonomously and dynamically alter its behavior during mission execution to improve the likelihood of a successful outcome.
Satellite Tracking with Optical Telescopes
Currently, the United States Air Force (USAF) is tasked with tracking approximately 17,000 man-made satellites using 29 sites that make up the Space Surveillance Network (SSN). However, the number of objects in space is increasing at an exponential rate making it very difficult for the current SSN to sufficiently accomplish this mission. To meet with future demand, the USAF has recently become interested in exploring the use of low-cost optical telescopes to supplement the existing SSN. Adding these new assets has the potential to improve both the number and the fidelity of tracks the network can produce, but these sensors have additional constraints for when they are appropriate to be used including the time of day, weather conditions, and other unforeseen outages. One way to compensate for these constraints is to deploy these sensors in large numbers, but that presents a scaling issue in the tasking of targets to sensors if the system is to be utilized efficiently.
Working with the Air Force Research Labs (AFRL) Advanced Planning Concepts Branch (AFRL/RISC), we are investigating the application of distributed resource allocation protocols to the problem of tasking the SSN to track satellites.
Distributed Problem Solving
Distributed problem solving is a subfield of Multi-Agent Systems (MAS) that focuses on developing protocols that allow agents to create solutions to a shared problem. In the CNAS lab, we develop protocols that solve myriad distributed problems including:
- Distributed resource allocation
- Distributed constraint satisfaction
- Distributed constraint optimization
From this work, we have developed a number of best-of-the-breed hill-climbing and cooperative mediation-based protocols including:
- Asynchronous Partial Overlay (APO)
- Optimal Asynchronous Partial Overlay (OptAPO)
- Distributed Probabilistic Protocol (DPP)
- Numerous dynamic variants of existing protocols (DynDBA, DynDSA, DynAPO)
Most recently we have developed new theoretical methods for analyzing these protocols based on thermodynamic theory.
The FARM simulator is an extendible distributed simulation platform that is designed to run large scale experiments. It works by allowing the various components of the simulator (display, state maintenance, logging, analysis, agents) to be run on different machines to distribute the workload. FARM has been tested with 10,000s of agents running on over 30 computer systems. A number of domains have been implemented using the FARM including sensor networks, satellite tracking, UAV airspace deconfliction, and generalized constraint solving. The FARM also has the capability to managing batch runs for testing scenarios with different parameter settings.