Hierarchical planning using neural subgoal generation

Building a world model takes exponential computational costs in the dimension of the configuration space. Furthermore, complexity increases with the number of obstacles, which in real world applications usually is high. Conventional algorithms can not even cope with slowly changing environments. In order to plan complex trajectories, a system that plans hierarchically shows many advantages. In this article we report the results of hierarchical planning using the neural subgoal generation system. We show that meaningful subgoals can be produced for manipulators in an environment with obstacles. Opposite to many other approaches our system works (once trained) fast and remains adaptive