Autorouting using a parallel Dijkstra algorithm with embedded constraints

An optimal path algorithm in which the paths are subjected to turning constraints has been developed. The restriction which has been incorporated is that the next link in the path must not make an angle exceeding 45° in magnitude with the preceding link. The algorithm is a variation of Dijkstra in that it follows a similar inductive procedure. The difference is that this algorithm stores more information for each vertex at each step. There is a constrained optimal path with ⩽k links flowing into a node through each adjoining link. At the kth step, Dijkstra would keep track of only the best one. In this procedure, all constrained optimal paths flowing into the node are recorded. The constraints may not allow continuation toward the goal vertex through the best of all the paths. Recording the extra information permits selection from allowable paths. There exists a natural implementation of this procedure as a neural network, and as a massively parallel algorithm on an array processor