Papers Published

  1. Young, R., and Garg, D., Modeling, Simulation, and Characterization of Distributed Multi-Agent Systems, Proceedings of the 2nd International Multi-Conference on Complexity, Informatics, and Cybernetics (March, 2011), pp. 179-184 .
    (last updated on 2011/08/15)

    Abstract:
    The goal of this novel research effort was to address the impact that variations in a pre-selectable composition of a multi-agent system (MAS) can have on its resulting capabilities and performance while utilizing a composition-tunable potential-function control scheme. This was accomplished by first defining the litany of capabilities and performance characteristics of individual agents, and the intra-agent relationships between these. Next, various incarnations of the MAS were hypothesized where distinctions existed through variables such as agent types, payload types and capabilities, relative quantities, motion behaviors, and most notably, potential function control tuning parameters. These were extensively represented and tested both in simulation and in hardware experimentation resulting in rich data sets. Then, trade-off analyses were conducted to identify those factors demonstrating significant importance. These analyses provided insight to the control system and hardware design such that the core elements of the MAS itself could be orchestrated to provide the best efficiencies against comprehensive mission accomplishment. Finally, the analyses provided insight to mission planners such that they would be able to tailor the composition of a task-specific MAS in terms of cost and performance.

    At its infancy, the study of mobile robotic technologies centered on simple wheeled and legged vehicles with elementary sensors and articulation to accomplish tasks such as movement through an area strewn with obstacles to find a goal location. Initial thrusts centered on providing these individual mobile platforms better sensing capabilities including high-resolution imagers and LIDARs. These platforms offered increasing capabilities; however, quickly became very complex and costly. Recently, the research community is exploring MAS whereby the entity accomplishing tasks is composed of a number of heterogeneous or homogeneous mobile robotic platforms (agents) that behave as a system through coordinated, centralized or distributed control theory. These systems offer significant potential and advantages in that each agent can be a simpler, and therefore more robust and inexpensive, robot while the coordinated system can still accomplish complex missions. Additionally, the MAS construct offers inherent benefits including redundancy, spatial efficiency, and graceful degradation.

    For this research, the critical enabling thrust explored was control via potential functions leveraging state space representation. A novel implementation of potential function control is developed that utilizes tuning parameters generated from state space relativity. For example, a tuning parameter consisting of the relative charge ratios of the individual agents and state space occupancy map cells is manipulated to optimize agent dispersion, path planning efficiency, and mission completion times.

    The results of this research show how the census of agents, including quantity and composition, ultimately impacts the overall performance of the MAS. For example, increases in quantity result in a non-linearly fading increase in performance measured by time to complete a pre-defined mission. This non-linearity means that more is not always better, especially in consideration of logistic burdens that increase with additional agent quantities. Also, the potential-function control architecture proved particularly robust and capable when used in conjunction with state space representation. Manipulation of the relative charges associated with MAS components and occupancy mapping resulted in a 25 percent improvement in task completion criteria. Finally, the concept of "sensing opportunity" is postulated where the combined capability of a MAS as a function of its agents' sensing systems is very much impacted by each sensor's opportunity potential to collect information of value to the desired MAS behavior, and not necessarily the sum total raw capability of the combined sensor payloads.