Home | People | Research | Publications| Links | News | Contact   

Environment-Agent Interaction in Autonomous Networked Teams with Applications to Minimum-Time Coordinated Control of Multi-Agent Systems

Recent technological advances in the field of autonomous vehicles and networked control systems have resulted in a growing need for more flexible mission planning and execution architectures both for military and civilian applications alike, compared to the traditional single-platform monolithic paradigm. Deployment of networks of autonomous vehicles (agents) would potentially result in more robust mission planning and execution, with increased levels of autonomy, and for a large spectrum of applications ranging from environmental monitoring, distributed surveillance, resource allocation, tracking of multiple targets, etc, to mention a few. This "network-centric'' paradigm utilizes teams of autonomous vehicles (aerial, terrestrial and underwater) working in unison by exchanging information to plan and execute a common mission in a decentralized, yet cooperative manner. This new paradigm brings about a slew of challenging problems, ranging from persistent communication, coordinated task assignment and prioritization, distributed estimation, distributed decision making, etc. These challenges are even more pronounced for small size UAVs/MAVs, which have dynamic responses dominated by short time scales and a strong interaction with the environment (that is, the effect of winds, the presence of pop-up threats, etc). They have also limited onboard computational and power resources. The objective of this research is to develop strategies for teams of UAVs that take into account the interaction with the environment and to do so in a computationally efficient manner to allow on-board implementation.

Motion coordination and path-planning algorithms for small UAVs/MAVs are investigated using appropriate, state-dependent performance metrics that capture the interactions of the vehicle with the environment, the mission objectives, as well as the system theoretic attributes of the network. The concept of generalized Zermelo-Voronoi diagram plays a key role in the proposed approach. In contrast to standard Voronoi decompositions that are based on Euclidean distance, generalized (Zermelo) Voronoi diagrams capture much better the dynamics of the state of the agents in the network (e.g., estimated-time-of-arrival). Several problems will be investigated in this context: First, computationally simple and efficient, distributed algorithms for the construction of generalized Voronoi diagrams with respect to suitable generalized metrics, are developed. These generalized Voronoi partitions can be tailored to the particular mission at hand. Second, and with the help of these diagrams, the proximity relations between different agents and/or between the agents and a set of targets can be characterized. The approach allows for the development of decentralized and distributed control strategies for a wide spectrum of applications, ranging from environmental monitoring, distributed surveillance, multiple target allocation, coordinated target intercept and attack, landing site selection, etc.


This project is supported by NSF.

Selected Publications





Home | People | Research | Publications| Links | Contact