On probability collectives for distributed control allocation

Stochastic optimization techniques by definition offer an element of randomness on the search trajectory for the best solution to an objective, and can be leveraged to gain confidence over traditional deterministic approaches. A novel method, Probability Collectives (PC) of Collective Intelligence (COIN), performs this type of stochastic search directly in the probability domain rather than in the variable domain, and enables a parallel processing architecture relevant to a distributed flight control system. Cooperating agents learn a local optimal strategy to minimize a global system cost. In this paper, we define these agents as individual aircraft effectors working to zero an acceleration error for the rigid body stability and control application. We present our initial acclimation to the PC framework with two forms of the method, delayed and immediate sampling, first in a simplified example setting with a two dimensional nonlinear objective, and finally in recent work for distributed control allocation of a hypothetical air vehicle with five effectors.