Continuous online task planning and plan acting in a deterministic robot environment

Abstract

Abstract In a deterministic environment, continuity in planning robot tasks and acting upon the generated plans is essential. This aspect does not commit the planner to generate all the required actions in advance, however; instead, the robot planner can generate the necessary abstract actions, then make the actor responsible for refining these actions into commands. Such commands are executed directly by the robot platform. In this paper, the planning and acting framework, known as RosPlanAct, is modified by including a temporal model that deals with the time interval for each action in the plan. Thus, a chronicle is associated with the abstract actions during the planning phase to act as a base for refining these actions into more detailed activities. The extended framework also has the ability to postpone the execution of any action that requires more details and to activate the monitoring function to compare what was observed with predicted results. This framework was tested in simulation and in real environments. The simulation environment was ROS, with a smart robot car used as a real testing environment. The results showed high performance in accomplishing the specified tasks with an increasing rate of success in executing the actions in the determined period of time and a low rate of failed execution. Throughout this paper, the actor was supported by being given the ability to avoid failure by online interactions between planning and acting sections in order to trigger the executing platform to perform the given task in a deterministic environment. The robot can thus refine the actions in the plan, re-planning as required in order to handle events arising.