Development of an integrated multi-robot task and motion planning system
Visualitza/Obre
Estadístiques de LA Referencia / Recolecta
Inclou dades d'ús des de 2022
Cita com:
hdl:2117/371636
Tipus de documentProjecte Final de Màster Oficial
Data2022-07-28
Condicions d'accésAccés obert
Llevat que s'hi indiqui el contrari, els
continguts d'aquesta obra estan subjectes a la llicència de Creative Commons
:
Reconeixement-NoComercial-CompartirIgual 3.0 Espanya
Abstract
The objective of this project is to develop an integrated framework for the automatic execution of complex Task and Motion planning problems in a multi-robot environment. In dynamic environments, such as the introduction of a human in the loop, the system cannot rely solely on an original plan. Possible disruptions occurring in the environment mid-execution, like an object being placed where it should not be by the human, or an object falling on its side, require a way to react to them and obtain a new plan in order to achieve the goal. The chosen tool to center the framework around is Behavior Trees (BTs). BTs have been chosen due to their ease of use, their modularity, and their portability. Each of the steps that are executed in the plan is represented as an Action Node. Through the use of Control Nodes, the order in which the different branches of the tree are executed can be dictated. The use of Reactive Control nodes allows for the system to react to the changes in the environment as soon as they occur. The framework begins by executing an original plan which has been planned offline based on the initial state of the environment. Once the BT executor is run, the system will check that the environment has not deviated from the original. If it has, a replan is triggered, and the BT will be rewritten mid-execution with the updated Task and Motion plan. Good monitoring of the system state is required in order to react to disruptions at either the geometric or the symbolic level. This led to the development of the State Module. Two different States are defined: the Expected State collects all the information from the original plan and represents the state the system should be in, while the Observed State uses the sensors in the perception module to sense the state the system is in. A mismatch between the two signifies that the original plan is no longer valid, and a replan is required. The robot chosen to perform most of the testing is a dual-armed 7 Degree of Freedom ABB YuMi. The YuMi robot does not offer a ROS interface, which would make it impossible to integrate the robot in the rest of the framework, which is based on ROS. For this reason, a ROS interface for the YuMi has been developed, which serves as a module for other robots and their ROS interfaces if they have to be integrated in the framework. Part of the framework involved converting the output of the motion planners, which are a series of waypoints in joint space, to a trajectory with the temporal component for the robot to perform. A new methodology named Waypint, short for "Waypoint Interpolation", has been developed, with the goal of generating efficient and robust trajectories considering the mechanical constraints of robotic manipulators. The experiments done to evaluate the framework proved its efficacy when purposely generating changes in the environment that required the system to perform a replan. Finally, the shortcomings of the implementation and potential avenues for improvement which remain as future work are discussed.
Fitxers | Descripció | Mida | Format | Visualitza |
---|---|---|---|---|
tfm-oriol-ruiz.pdf | 11,71Mb | Visualitza/Obre |