Busca avançada
Ano de início
Entree
(Referência obtida automaticamente do Web of Science, por meio da informação sobre o financiamento pela FAPESP e o número do processo correspondente, incluída na publicação pelos autores.)

A Faddeev Systolic Array for EKF-SLAM and its Arithmetic Data Representation Impact on FPGA

Texto completo
Autor(es):
Rosa, Leandro de Souza [1] ; Dasu, Aravind [1, 2] ; Diniz, Pedro C. [2] ; Bonato, Vanderlei [1]
Número total de Autores: 4
Afiliação do(s) autor(es):
[1] Univ Sao Paulo, Inst Math & Comp Sci, Sao Paulo - Brazil
[2] Univ Southern Calif, Inst Informat Sci, Los Angeles, CA 90007 - USA
Número total de Afiliações: 2
Tipo de documento: Artigo Científico
Fonte: Journal of Signal Processing Systems for Signal Image and Video Technology; v. 90, n. 3, p. 357-369, MAR 2018.
Citações Web of Science: 0
Resumo

The Extended Kalman Filter (EKF) computation is a core task for the simultaneous localization and mapping (SLAM) problem in autonomous mobile robots. The SLAM problem involves operations over high dimension data sets, requiring high throughput and performance, given the real-time nature of the robotics, control-decision algorithm this task is a part of. The lightweight and power restricted computing environments in mobile robotics requires customized processing systems such as Field-Programmable Gate Arrays (FPGAs). This work presents an arithmetic precision analysis and a Faddeev algorithm to calculate the Schur's Complement hardware architecture implementation for the EKF-SLAM using a Systolic Array (SA). While it is widely believed that fixed-point implementations of arithmetic operations lead to area and performance benefits on FPGAs, the results in this article reveal that each Processing Element (PE) in the SA consumes 25% more logic and about 30% more register resources for the fixed-point 13.23 representation than if using the IEEE-754 single precision floating-point format. In addition, for FPGA devices with hardware support for key components of floating-point computations, a single PE floating-point implementation can achieve a maximum frequency up to 50% higher than a corresponding fixed-point implementation for the same relative numeric errors. (AU)

Processo FAPESP: 12/20224-8 - Um hardware para o EKF-SLAM em ponto fixo para robótica embarcada
Beneficiário:Leandro de Souza Rosa
Linha de fomento: Bolsas no Exterior - Estágio de Pesquisa - Iniciação Científica