Recent advances in batteries, motors and sensors allow consideration of tasks such as mapping,
exploration, inspection, and transport-- previously only the domain of ground vehicles—to be now
conducted by aerial vehicles. While we have been able to automate flight scripted to stay clear of the
environment for some time, operation in complex and unmapped environments is new. We might draw
inspiration from the avian world to enable navigation from one point to another, but there is a question
of how to tie these elements into what we really want these machines to do-- map, explore, and team.
This is especially hard given the complexity in control and the constraints on computing and sensing.
In this talk, I will discuss how we might organize familiar capabilities in perception and motion planning
to enable safe flight at high speeds in complex 3D environments. Spoiler Alert: we succeed by violating
sacrosanct norms of filtering, without needing to resort to Monte Carlo methods. I will provide
examples of implementations on aircraft ranging in scale from submeter drones to 10 meters (full-scale
helicopters).