FPGA-based Accelerator for Convolutional Neural Network Application in Mobile Robotics

• Lucas F. R. Mazzetto Unesp
• José E. C. Castanho Unesp

Resumo

This paper presents the development of a Convolutional Neural Network accelerator based on a Field Programmable Gate Array (FPGA) for the steering control application in a mobile robot. The FPGA-based implementation was benchmarked against an equivalent implementation using a general-purpose processor. The FPGA outperforms the general-purpose processor by significantly reducing processing time. However, the quantization process introduces increased estimation error, and the concurrent execution of the processor for data preprocessing in the accelerator implementation leads to higher power consumption. While the proposed solution is specific to the robot steering control problem, the general insights can be extrapolated to other applications that require high computing performance within power constraints.

Palavras-chave: FPGA, Hardware Accelerator, Convolutional Neural Networks, Mobile Robotics, Embedded Systems