PPGCC/CCET

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE: MATHEUS CHAVES MENEZES
DATA: 22/03/2019
HORA: 09:00
LOCAL: Auditório do DEINF
TÍTULO: BRAIN-TO-BRAIN MAPPING: AN APPROACH TO SHARE NEURAL INFORMATION ON RATSLAM
PALAVRAS-CHAVES: Robotics, SLAM, RatSLAM
PÁGINAS: 59
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
RESUMO: Several robotic applications are better performed by systems with multiple robots rather than only one, for example, to explore large areas in time-critical post-disaster search and rescue missions. These advantages can be due to the division of activities, cost and time reduction. Simultaneous Localization and Mapping (SLAM) plays a central role in exploring unknown environments. RatSLAM, which is based on the navigation system present in the hippocampus of rodents’ brain, has been widely used on video-based SLAM applications. In RatSLAM, neural information is defined as experiences, which associates characteristics of the environment and movement in a unique representation on the map. This work presents an approach to share neural information on RatSLAM, named Brainto-brain mapping, in which experience from partial maps are shared by various robots to cooperatively construct a map of the entire environment. The first step to share neural information is to connect different instances of RatSLAM through a merge mechanism, specific for RatSLAM. To perform the merge, it is necessary that the robots pass through at least a common place among them and acquire the same experience about the common place. The merge enables all robots to know about their experiences (pose cells, local view e experience map) in a shared structure. Thus, the exploration robots can reuse learned experiences over the environment to improve their map procedure e.g. a robot can correct part of the map of another robot, while using shared information to improve its own map performing loop closure. Three experiments of different environments were carry out to validate the new approach: a simulated environment, a research lab, and a dataset used to validate the original work of RatSLAM. The results has showed that the final map built by robots with shared experience is visually similar (but not identical) to one built by one robot performing the same mapping task individually, i.e. without sharing information.
MEMBROS DA BANCA:
Presidente - 1091306 - ALEXANDRE CESAR MUNIZ DE OLIVEIRA
Interno - 2044484 - AREOLINO DE ALMEIDA NETO
Externo à Instituição - EDISON PIGNATON DE FREITAS - UFRGS
Interno - 2319041 - PAULO ROGERIO DE ALMEIDA RIBEIRO