A complete task planning and execution framework for scenarios with mobile manipulators

  • Ein vollständiges Framework zur Aufgabenplanung und -ausführung für Szenarien mit mobilen Manipulatoren

Bezrucav, Stefan-Octavian; Corves, Burkhard (Thesis advisor); Nitsch, Verena (Thesis advisor)

Aachen : RWTH Aachen University (2023)
Dissertation / PhD Thesis

In: Institut für Getriebetechnik, Maschinendynamik und Robotik
Page(s)/Article-Nr.: 1 Online-Ressource : Illustrationen, Diagramme

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023

Abstract

The deployment of robotic solutions as part of teams with humans still needs to be adapted on a large scale. One of the reasons is the lack of complete frameworks that combine task planning and execution methods. The planning methods deliberate about the process steps that must be carried out and distribute the actions among the members of the team such that the team achieves optimally set goals. Following, the task execution procedures trigger and supervise the execution of the planned actions. This thesis presents a self-contained task planning and task execution framework that enables dynamic human-robot cooperation processes. The proposed framework implements a high-level sensing-planning-execution control loop that extends a state-of-the-art framework with novel planning models and features. A planning situation is an abstract representation of a real-world situation in which a set of changes modify the current state of the world to a new state in which set goals are reached. Such planning situations are formulated as planning problems that can be solved. The solution is a plan with the actions that must be carried out to impose the wanted changes on the world. The first contribution of this thesis is the generic formulation of planning problems for scenarios involving teams of cooperating humans and robots. The formulation is done in the Planning Domain Definition Language (PDDL), the standard language for automated planning methods. The class of automated planning methods is selected due to its capability to cope with variations of planning problems easily. The second contribution of this thesis is the extended framework that enables, besides the planning process, a resilient and robust execution of the planned actions. Among others, these improvements are achieved by a novel Supervisor that triggers the execution of the actions and handles re-plan requests, the Action Interface Manager that implements the interfaces between the abstract formulation of the planned actions and real-world execution modules, and the fast re-planning approach. Test campaigns with simulated and real-world laboratory scenarios validate both the planning models and the complete planning and execution framework. In all scenarios, a team of one human and one robot successfully executes at least fifteen different process steps. Besides the complete execution, the results show that the new framework handles unforeseen changes (e.g., execution errors) and is able to reduce waiting and total execution times significantly. Thus, the framework is prepared for deployment in real applications.

Institutions

  • Chair and Institute of Mechanism Theory, Machine Dynamics and Robotics [411910]

Identifier

Downloads