An information fusion approach to missile guidance

We address the problem of missile guidance in the presence of noisy measurements. State-of-the-art interception missiles derive their superior performance mainly from highly advanced aerodynamic configuration design, agile seekers and highly efficient fragmentation warheads, but use only relatively basic information on the state of the target. Recent advances in onboard computing and imaging technologies render the expansion of this information base feasible, thus enabling the use of advanced guidance laws. In this plenary talk we focus on an integrated, information fusion approach to missile guidance, which facilitates synergy between the target estimation and guidance functions of the missile autopilot. In the first part of the talk we examine the advantages of fusing classical line-of-sight observations with target orientation (attitude) observations, obtained by using computer vision techniques on visual measurements acquired by high-resolution imaging sensors. We show that, by properly exploiting the various sources of information, and using the target´s bank-to-turn mathematical model, the missile´s closed-loop interception performance can be vastly improved even when using an off-the-shelf differential game-based guidance law, which is derived based on an unrealistic perfect information assumption. In the second part of the talk we make a bolder step in the direction of estimation/guidance fusion. Existing missile guidance law design methods traditionally assume the validity of the separation theorem. While this assumption permits separate designs of the optimal estimation and guidance (control) functions, it has never been proven valid in realistic (nonlinear, non-Gaussian) interception scenarios. In such cases, only a generalized separation theorem, first introduced by Witsenhausen in 1971, may be applied. In this plenary talk we present a geometry-based approach to fusion of estimation and guidance, that is consistent with the generalized separation theo- rem. A nonlinear, non-Gaussian numerical study is presented, that demonstrates the performance of the proposed methodology in a 3-D realistic engagement scenario with partial information.