Simultaneous Non-Myopic Optimization of UAV Guidance and Camera Gimbal Control for Target Tracking

Abstract

Inexpensive lightweight gimbal-mounted cameras have become a viable option for replacing fixed camera systems mounted to a small UAVs for target tracking applications. The gimballed camera adds degrees of freedom, expanding the limited field of view of the camera. This is advantageous but the additional degrees of freedom equate to augmenting the control space to include the gimbal controls. While the addition of gimbal controls is not an insurmountable obstacle, the slew rate of the gimbal axis motors is often limited due to hardware performance capabilities or system definitions. This imposes constraints on the controls. Target-tracking scenarios often include sparsely distributed targets in which long-term tracking performance greatly benefits from non-myopic control. Under these considerations, the gimbal control constraints affect the reachable set of camera-aiming angles, in turn directly impacting the UAV guidance. To maximize the benefits provided by the gimbal, we formulate the problem of simultaneous UAV and camera gimbal control as a partially observable Markov decision process (POMDP), applying a Q-value approximation technique called nominal belief-state optimization (NBO). We combine a receding-horizon objective function with a heuristic expected cost-to-go (HECTG) to approximate target-tracking performance over an extended horizon at a reduced computational cost.