Semi-plenary multi-agent control with multi-resolution sensing over networks: A semidefinite programming approach

This semi-plenary is about the systematic design of networked multi-agent systems at fast time scales, and finds its basis in mathematically grounded questions about collaboration and control in such dynamical systems. It is inspired from a practical perspective by recent advances in sensing, computing and networking hardware that make reconfigurable multi-agent systems both technologically and economically feasible on a widespread scale. Examples of such systems include satellite formations, consumer robotics, and more generally networked subsystems of mobile sensors and actuators that cooperate to achieve some aggregate functionality. There are many shared design challenges associated with these types of systems, and the talk will focus on three specific connected issues: distributed control over graphs; tools for switched hybrid dynamics; and network latency and multi-resolution sensing for control. The technical setting for the talk is the framework of convex optimization, and the results presented lead directly to implementable analysis and design algorithms via semidefinite programs (SDPs). A portion of the presentation will involve describing the experimental multi-vehicle testbed, HoTDeC, which consists of wirelessly linked miniature hovercraft and wheeled robots capable of onboard multi-modal sensing, and interaction with the Internet and a vision-based sensor network.