Robot control can be made easier, if the task is divided into subtasks. Each subtask is equal to a goal. After formalizing the goals into machine detectable events, it's possible to track the robot's execution in comparison to the model.
Apart from simple monitoring the steps of the robot, already defined subgoals can be utilized in a motion planner. This helps to improve the search efficiency in Reinforcement learning situations in which the reward for an artificial agent is unknown.
Another advantage of a goal hierarchy is to predict the outcome of future model states. The anticipated goals are similar to a mental imaginary which is guiding the movement of the robot.
- Zeng, Junjie, et al. "Combining Subgoal Graphs with Reinforcement Learning to Build a Rational Pathfinder." Applied Sciences 9.2 (2019): 323.
- Krichmar, Jeffrey L., et al. "Advantage of prediction and mental imagery for goal-directed behaviour in agents and robots." Cognitive Computation and Systems (2019).