Closed-Loop Controls for Differential Games Using a Gradient and a Differential Dynamic Programming Method.

摘要

This thesis investigates the use of a gradient method and a combined gradient-differential dynamic programming (DDP) method to generate closed-loop controls for intercept problems formulated as differential games. The gradient method is applied to a planar motion pursuit-evasion game. The trajectory obtained compares favorably with that obtained using analytic expressions for closed-loop controls. The gradient method is applied to a three dimensional interceptor-penetrator game with simplified dynamics on a real-time basis. A combined gradient-DDP algorithm is applied to this problem but not on a real-time basis. The DDP portion of this combined control law was found to be unstable. The results obtained indicate that a gradient based scheme, because of its numerical stability and ability to rapidly converge to the vicinity of the optimum, may be used to generate an effective near-optimal closed-loop control law for some problems. (Author)