GALCIT Colloquium

We will discuss an ambitious project which proposes a low-cost balloon observation system for sustained (in time), broadly distributed (in space), in-situ (between 1-8km altitude), real-time measurement (from data acquisition to NCAR within 20 minutes) of hurricane development. The high density (in both space and time) measurements from such a robotic vehicle swarm (100 or more vehicles) will be invaluable in significantly improving our ability to estimate and forecast such extreme and dangerous atmospheric events.

Challenges we will discuss in this over-arching problem, which is of acute societal relevance, include:

• the design of small (3 kg, 5.5 m^3 at 8 km altitude), inexpensive (< $ 2k), robust, sensor-laden, buoyancy-controlled balloons that don't accumulate ice, and are deployable from the launch chutes (13 cm diameter x 91 cm long) of existing NOAA P3 aircraft,
• the implementation of a self-reconfiguring Mobile Ad hoc Network (MANET) over the balloon swarm, leveraging ultra-low-power cellphone or IoT radios communicating in the UHF band over (typically) 10 to 30 km distances, which will be used to relay messages to/from a satellite uplink operating in the core of the hurricane, and
• the development of efficient hierarchical systems-level control algorithms to autonomously coordinate the vertical motion of the balloons in the swarm to move them with the hurricane, simultaneously achieving both good coverage and good connectivity while minimizing the control energy used, leveraging the strong vertical stratification of the horizontal winds to distribute the balloons in the desired fashion horizontally. We will discuss two distinct classes of coordination algorithms for this problem, in addition to their hierarchical combination:
  • a (centralized) Model Predictive Control (MPC) strategy for coordinating the large-scale balloon distribution, leveraging coarse flowfield forecasts developed with the cutting-edge hurricane Weather and Research Forecasting (WRF) code developed at NOAA, and
  • a (decentralized) hysteretic Three-Level Control strategy for rejecting the smaller-scale disturbances that arise due to unresolved turbulent flowfield fluctuations, as well as a refined disturbance rejection strategy based on a "Drunken Sailor" random walk model for the ballon velocity (rather than its position), with a concomitant realistic ω^-2 energy spectrum, resulting in a Spontaneous Singularity in the optimized control distribution (which is highly practical as the control is in fact turned off over most of phase space).

The presentation can be found at:

http://renaissance.ucsd.edu/courses/mae293/Hurricanes_Caltech.pdf