Dynamic Data-driven Autonomous Mapping

July 31, 2013

Admit it...every time you looked at a smokestack you've wondered what it does to the environment. You wished you could map the structure of the plume and follow the puffs. Small UAS' can be made to do that, and much more, using a dynamic data-driven mapping paradigm. There's much more sophisticated language for the intuitive concept of DDDAS, but the essence of the idea is to sample, then resample based on information that your sensors have gathered. This symbiosis is where the magic is. Examples follow of a  Markov Chain Monte-Carlo adaptive sampling, but make sure you are in a glide (and if that's not possible, maintain trim!) on the sampling flight path... Here are some pix for you!


This dynamic data-driven application looks like this. Into the domain the aircraft enters a glide and a variometer detects the plume. As the plume is encountered, an analysis is made through GP, and the domain is resampled so that the paths are higher in density in a vicinity of encountered measurements. That is, any path that improves fidelity of map to sampled points is immediately accepted, but ones that do not are accepted with a probability -- this is a classical Metropolis-Hastings approach. Eventually the paths and maps reach equilibrium and represent the underlying plume, here for a single "layer".  The glide path ensemble can be used to recover the kinetic energy structure and that is such fun to see!