Neural network guided search control in partial order planning

Planning programs integrated with neural network (NN) were investigated to develop efficient search control methods in assisting planner's search for solution. Partial plan parameters that would likely evolve as a solution is entered to a planning program. The program was then modified to automatically produce vector for every partial plan used in its search process. The modified planner runs on a set of problem and generates two sets of input vector, one for network training and the other for testing. The training set of input vectors were used in designing and training a NN. The modified version of the partial order planner that incorporates the threshold function representing the successfully trained NN is developed to guide planners.