Stochastic Optimal Control of Guided Parafoil Systems

  • Picture of GPU flight test
Guided parafoil systems provide a unique capability for precision airdrop that is highly attractive for a variety of missions.  Currently, the military and other civilian organizations are interested in developing autonomous guided parafoils in an array of sizes.  These next generation autonomous parafoils are expected to provide unprecedented landing accuracy even in unfavorable environmental conditions using advanced sensor technology and optimal control algorithms. While parafoils, as a class of unmanned systems, are fundamentally simple to control due to their stability and benign dynamics, several factors make the specific task of precision parafoil landing extremely challenging.  First, parafoils are underactuated systems in that they typically employ only lateral control through asymmetric brake deflection with longitudinal control for guidance being extremely limited.  Secondly, low airspeed and the absence of thrust make parafoils extremely sensitive to winds, which can be highly unpredictable.  Winds are often a primary driver in landing point accuracy, and it is a common occurrence that wind speed approaches the vehicle airspeed such that advancing upwind flight is difficult.

This work is focused on developing a new type of parafoil terminal guidance algorithm that addresses the need for robust guidance by exploiting massively-parallel processing.  The terminal guidance planning method we have proposed relies on online, massively-parallel Monte Carlo simulation to predict impact point variance resulting from unknown or changing wind conditions.  These Monte Carlo simulations are designed to be performed on embedded graphics processing units, which are increasingly being used for general purpose computation.  In our previous work, we exercised the GPU-based control system to explore precision landing in drop zones constrained by terrain obstructions.  Results showed that the guidance system shaped the terminal trajectories successfully so as to minimize the probability of collision with terrain given wind disturbances.  This capability is highly attractive for challenging drop zones that may be surrounded by mountainous terrain.

Given promising results in simulation, we recently flew a custom autonomous parafoil equipped with an embedded GPU to demonstrate performance in flight experiments.  These successful flight tests, the first of their kind to incorporate onboard massively-parallel processing, have demonstrated the viability of stochastic control on fielded parafoil systems.

Plot of standard guidance Plot of stochastic guidance Image of stochastic electronics setup

References:

  1. J. Rogers, N. Slegers, “Robust Parafoil Terminal Guidance Using Massively Parallel Processing,” Journal of Guidance, Control, and Dynamics, Vol. 36, No. 5, September-October 2013, pp. 1336-1345.
  2. N. Slegers, J. Rogers, “Terminal Guidance for Complex Dropzones Using Massively Parallel Processing,” AIAA Aerodynamic Decelerator Systems Technology Conference and Exhibit, Daytona Beach, FL, March 25-28, 2013.
  3. J. Rogers, “Massively-Parallel Stochastic Control and Automation: A New Paradigm in Robotics,” GPU Technology Conference, San Jose, CA, March 24-27, 2014.
  4. http://www.hpcwire.com/2014/01/24/scientist-conducts-first-cuda-powered-flight/