OATAO - Open Archive Toulouse Archive Ouverte Open Access Week

An ontology-based approach towards coupling task and path planning for the simulation of manipulation tasks

Zhao, Yingshen. An ontology-based approach towards coupling task and path planning for the simulation of manipulation tasks. PhD, Robotique et Informatique, Institut National Polytechnique de Toulouse, 2019

[img]
Preview
(Document in English)

PDF (Author's version) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
9MB

Abstract

This work deals with the simulation and the validation of complex manipulation tasks under strong geometric constraints in virtual environments. The targeted applications relate to the industry 4.0 framework; as up-to-date products are more and more integrated and the economic competition increases, industrial companies express the need to validate, from design stage on, not only the static CAD models of their products but also the tasks (e.g., assembly or maintenance) related to their Product Lifecycle Management (PLM). The scientific community looked at this issue from two points of view: - Task planning decomposes a manipulation task to be realized into a sequence of primitive actions (i.e., a task plan) - Path planning computes collision-free trajectories, notably for the manipulated objects. It traditionally uses purely geometric data, which leads to classical limitations (possible high computational processing times, low relevance of the proposed trajectory concerning the task to be performed, or failure); recent works have shown the interest of using higher abstraction level data. Joint task and path planning approaches found in the literature usually perform a classical task planning step, and then check out the feasibility of path planning requests associated with the primitive actions of this task plan. The link between task and path planning has to be improved, notably because of the lack of loopback between the path planning level and the task planning level: - The path planning information used to question the task plan is usually limited to the motion feasibility where richer information such as the relevance or the complexity of the proposed path would be needed; - path planning queries traditionally use purely geometric data and/or “blind” path planning methods (e.g., RRT), and no task-related information is used at the path planning level Our work focuses on using task level information at the path planning level. The path planning algorithm considered is RRT; we chose such a probabilistic algorithm because we consider path planning for the simulation and the validation of complex tasks under strong geometric constraints. We propose an ontology-based approach to use task level information to specify path planning queries for the primitive actions of a task plan. First, we propose an ontology to conceptualize the knowledge about the 3D environment in which the simulated task takes place. The environment where the simulated task takes place is considered as a closed part of 3D Cartesian space cluttered with mobile/fixed obstacles (considered as rigid bodies). It is represented by a digital model relying on a multilayer architecture involving semantic, topologic and geometric data. The originality of the proposed ontology lies in the fact that it conceptualizes heterogeneous knowledge about both the obstacles and the free space models. Second, we exploit this ontology to automatically generate a path planning query associated to each given primitive action of a task plan. Through a reasoning process involving the primitive actions instantiated in the ontology, we are able to infer the start and the goal configurations, as well as task-related geometric constraints. Finally, a multi-level path planner is called to generate the corresponding trajectory. The contributions of this work have been validated by full simulation of several manipulation tasks under strong geometric constraints. The results obtained demonstrate that using task-related information allows better control on the RRT path planning algorithm involved to check the motion feasibility for the primitive actions of a task plan, leading to lower computational time and more relevant trajectories for primitive actions.

Item Type:PhD Thesis
Uncontrolled Keywords:
Institution:Université de Toulouse > Institut National Polytechnique de Toulouse - Toulouse INP (FRANCE)
Laboratory name:
Research Director:
Archimède, Bernard and Fillatreau, Philippe
Statistics:download
Deposited On:05 May 2020 13:26

Repository Staff Only: item control page