Coordinated OAV Formations: Decomposition, Decentralization and Optimization

This work presents an overview of current efforts in the area of coordinated control of Organic Air Vehicles (OAVs) at Honeywell Laboratories within the context of formation flight. First, the dynamics of the OAV and its associated Guidance and Control (G&C) loops are briefly discussed. A description of the higher level decentralized optimizationbased scheme used to achieve formation flight for this class of vehicles then follows. Interest in coordinated control of Unmanned Air Vehicles (UAVs) has grown significantly in recent times. The main motivation is the wide range of military and civilian applications where teams of UAVs acting in a coordinated fashion could provide a low cost and efficient alternative to existing technology. Many of these applications require the UAVs to fly certain formations in order to accomplish their assigned tasks. Among these, distributed sensing applications are envisioned to be the most appealing. Such applications include Synthetic Aperture Radar (SAR) interferometry, phased antenna arrays, surveillance, damage assessment, reconnaissance and chemical or biological agent monitoring. These kinds of applications require the development of control system design techniques for large and tight formations. For our purposes, formation flight is a large control problem in which we seek to compute the inputs that drive the vehicles along trajectories that maintain relative positions as well as safe distances between each OAV pair. Optimal control problem formulation has been one of the more successful frameworks used to tackle such problems [1], [2], [3], [4], [5], [6], [7]. In this approach, the problem is formulated as a minimization of the error between OAV relative distances and desired displacements. Collision avoidance requirements can be easily included as additional constraints between each OAV pair in the optimal control problem. Centralized optimal or suboptimal approaches have been used in different studies [8], [5], [4]. However, as the number of vehicles increases, the solution of big, centralized, nonconvex optimization problems becomes prohibitive. This is true even when the most advanced optimization solvers and (over)simplified linear vehicle dynamics are used [9]. The main challenge is to formulate simpler decentralized problems which result in a formation behavior similar to the one obtained with a centralized approach. In this paper we present an overview of the current work on OAV formation flight carried out at the Honeywell Laboratories in Minneapolis. The OAV is a scalable autonomous ducted-fan vehicle which is intended to be used as a demonstrator vehicle for the U.S. Army’s Future Combat Systems (FCS) program. This vehicle is depicted in Figure 1(a). It exhibits a highly nonlinear and constrained multi-variable character. Because of its vertical take-off and landing (VTOL) capability, as well as its relatively small size, such a vehicle offers several potential tactical advantages to war-fighters of the future. This is especially true for platoonand other lower-level Reconnaissance, Surveillance, and Target Acquisition (RSTA). There are also numerous envisaged applications of smaller-scale versions of this vehicle for homeland security. Our technical approach to the problem renders the complex task of multi-OAV formation control/reconfiguration tractable via hierarchical decomposition. In such a decomposition, the lower level is made up of the OAV dynamics equipped with efficient guidance and control loops. At the higher level, the controlled OAV is represented with sufficient fidelity as a constrained multi-input, multi-output (MIMO) linear system. For this class of systems, a decentralized optimization-based control framework is then developed to achieve formation flight, formation reconfiguration and other cooperative tasks. The first step in the design procedure is to develop a model of the OAV. This is accomplished by (conceptually) breaking it up into its constituent parts, developing models for each, and then putting the pieces back together. These building blocks are shown in Figure 1(b). The vehicle translates and rotates by modulating thrust and prop wash along its axes via the deflection of specific vanes sets. The four sets of vanes are left and right pitch vanes; and forward and aft roll vanes. In the dynamic model, the inputs are the vane deflections and the thrust along the prop axis. Wind tunnel experiments are carried out to obtain aerodynamic and propulsion tables that provide a mapping from these vane deflections/thrusts to the forces and moments that act on the vehicle. Rigid body dynamics are obtained from the physics of the vehicle. The second step is to design low-level guidance and control loops for the vehicle. Nonlinear control of the ‘inner loop’ (i.e. attitude and rate loops) and the ‘outer loop’ (position and velocity) loops are accomplished via nonlinear dynamic