AbstractThis communication presents computational results with a variant of differential dynamic programming. The class of methods is motivated by quasi‐Newton methods and constructs quadratic approximations by using stagewise gradients. We compare two methods of updating the approximations: the first is based on the symmetric Broyden update and the second is a sequential rank‐one/rank‐two update. Our computational results suggest that the latter is a practical and efficient algorithm for discrete‐time optimal
展开▼