Q-Learning of Spatial Actions for Hierarchical Planner of Cognitive Agents

In the paper, we consider the problem of the robotic movement inaccuracy. We suggest that clarifying the abstract actions of the behavior planner will help build more precise control of the robot. A multi-agent planner for the synthesis of the abstract actions with refinement for a two-dimensional movement task was proposed. We analyze the problem of the action execution by robots and present a way to solve navigation problems through the use of reinforcement learning and deep learning algorithms. This method made it possible to synthesize sets of atomic sub-actions for correcting the state of the robotic platform at each moment of time. We conducted a set of experiments in single and multiagent settings. The synthesis of sub-actions of several tasks formed a training set for the RL model, which was tested on a test example. All considered tasks consisted of moving robotic platforms across a map with obstacles and manipulating environmental objects.