Toward Real Time Autonomous Robotics Through Integration of Hierarchical Goal Network Planning and Low Level Control Libraries

Automated planning has become an increasingly influential area of research in the realm of artificial intelligence. Task based planning algorithms provide a number of advantages including the ease of human readability when creating mission length plans. However, task based planning algorithms are rarely implement on real world robotic systems because they require additional domain specific knowledge to define tasks and are generally not as flexible as other planning techniques. This paper documents work to integrate a hierarchical goal network planning algorithm with low level path planning. The system utilizes the Goal Decomposition with Landmarks (GoDeL) planner for plan generation at an abstract level and the Searched Based Planning Library (SBPL) for low level control. The system is used to direct a robot through an office setting within a simulation environment. We then discuss incorporating an ”in the now” approach to the GoDeL algorithm to make the system more robust to a dynamic environment. The resulting algorithm is more suited for use in real time applications such as autonomous robotics.