Português English
Contato

Proposta de Tese de Mathias Fassini Mantelli


Detalhes do Evento


Aluno: Mathias Fassini Mantelli
Orientadora: Profa. Dra. Mariana Luderitz Kolberg Fernandes
Coorientador: Prof. Dr. Renan de Queiroz Maffei
Título: Exploiting semantic information in indoor environments
Linha de Pesquisa:
Planejamento, Sistemas Multiagentes e Robótica

Data:  10/12/2021
Horário: 14h30min.
Esta banca ocorrerá excepcionalmente de forma totalmente remota. Interessados em assistir a defesa poderão acessar a sala virtual através do link: https://meet.google.com/trr-vscf-xfe

Banca Examinadora:
– Prof. Dr. Joel Luís Carbonera (UFRGS)
– Prof. Dr. Roseli Aparecida Francelin Romero (ICMC/USP)
– Prof. Dr. Vitor Augusto Machado Jorge (UFABC)

Presidente da Banca: Profa. Dra. Mariana Luderitz Kolberg Fernandes

Abstract: Nowadays, the mobile robotics research community deals with different high-level asks that require the robot to manipulate or interact with objects which are not in the robot’s field of view or in an unexplored region of the environment. To find an object in unknown environments, the robot needs to look for it while gaining information about the environment and making decisions in real-time, known as the object search (OS) problem. The research community has proposed different approaches for dealing with the OS problem, relying on the objects’ color, 3D shape, or its surroundings as visual cues to guide the search. However, this geometric information (i.e., color or size) limits the robot’s perception and, consequently, the robot’s performance during the search. Therefore, we proposed exploiting the advantages of semantic information inferred from two sources abundant in human-populated environments: text and dynamic agents. First, we propose an OS system for unknown indoor environments that relies on semantic information inferred from texts found in the environment, aiming to find a target door label. The use of semantic information in this scenario allowed the robot to reduce the search costs by avoiding not promising regions to contain the target door label. Our semantic planner reasons over the numbers detected from door labels to decide either to continue the search towards unknown parts of the environment or carefully search in the already known parts. Second, we present another OS system based on semantic information inferred from different objects’ position over time in the environment. Composed by two modes, our system first gathers data from the objects’ placement by executing its recording mode. This data is later used when the robot executes the request mode to search for the target object. Both systems were evaluated in different environments and compared against other OS approaches in simulated and real scenarios. The results support our systems’ efficiency and demonstrate the improvement in the searching performance with the aid of semantic information.

Keywords: Mobile robotics. Object search. Semantic information. Robotics perception. Indoor environments.