Flyer: Perception-Aware Navigation: From Trajectory Optimization To Heterogeneous Robot Teams

State estimation is one of the core components of all autonomous systems. For size, weight,
and power (SWaP) constrained robots, it is often performed with a special assortment of
sensors tailored to the size and power capacities of the agent. If there is enough perceptual
information in the environment, the robot can extract accurate state estimates. However,
when such information is scarce, there is a fundamental barrier. An agent must either act
in a way cognizant of such circumstances, or we must resort to alternate solutions
altogether. In this talk, I will first consider the problem of minimizing the traversal time of
a given geometric path by a vision-driven micro aerial vehicle that has to maintain accurate
state estimates at all times. In the second part, I will shift the focus towards enabling
autonomous agents to navigate perceptually-degraded environments. In particular, I will
look at how members of robot teams can leverage each other to localize accurately. We
shall delve into the underlying optimization problems, showing that suitably positioning
perceptually-advantaged members of the team can reduce the localization uncertainty of
perceptually-disadvantaged agents by up to ninety percent, as well as showing what cannot
be achieved regardless of which optimization algorithm we use.