摘要
基于现场可编程门阵列的片上系统在边缘端人工智能应用中具有独特的优势。其神经网络推理加速由软核实现,可随着人工智能技术的发展更新硬件加速器而不用更换芯片方案,同时,FPGA侧也可针对具体应用定制其他专用的硬件加速核。本文使用Xilinx KV260开发板和Vitis AI工具链,在FPGA的SoC上实现硬件加速神经网络推理的基于示教学习的自动驾驶小车系统。
Abstract
SoCs with embedded FPGA have a unique property in edge AI applications.Its neural network inferencing accelerator is implemented with soft IP cores on the FPGA side and can be upgraded during the product lifecycle,and its FPGA resources can be used to implement other application-specific hardware accelerators.This paper uses Xilinx KV260 development board and Vitis AI tool chain to realize hardware accelerated neural network reasoning based on demonstration learning automatic driving car system on SoC of FPGA.
关键词
边缘人工智能 /
现场可编程门阵列 /
Xilinx KV260
Key words
edge AI /
FPGA /
Xilinx KV260
郭传鈜, 王延葵, 毕盛, 董敏.
基于Xilinx KV260和卷积神经网络的自动驾驶小车*[J]. 集成电路与嵌入式系统. 2022, 22(10): 3-6
Guo Chuanhong, Wang Yankui, Bi Sheng, Dong Min.
Self-driving Model Car Based on Xilinx KV260 and Convolutional Neural Network[J]. Integrated Circuits and Embedded Systems. 2022, 22(10): 3-6
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 2022年45家国产AI芯片厂商调研分析报告[EB/OL].[202208].https://zhuanlan.zhihu.com/p/505961678.
[2] ROS简介[EB/OL].[202208].http://wiki.ros.org/cn/ROS.
[3] 郭传鈜.作品回顾｜基于视觉演示学习的自动驾驶小车[EB/OL]. [202208].https://mp.weixin.qq.com/s/_8n8PuGNQhQ6KLi_WXvFw.
[4] PyTorch教程以CIFAR10为例[EB/OL].[202208].https://blog.csdn.net/weixin_42468475/article/details/110565705.
[5] Using Vivado to Build the Hardware Design[EB/OL].[202208].https://xilinx.github.io/kriaappsdocs/main/build/html/docs/build_vivado_design.html.
[6] Recording and playing back data[EB/OL].[202208].http://wiki.ros.org/rosbag/Tutorials/Recording%20and%20playing%20back%20data.
[7] Reading messages from a bag file[EB/OL].[202208].http://wiki.ros.org/ROS/Tutorials/reading%20msgs%20from%20a%20bag%20file.
[8] PyTorch编写自定义数据集、数据加载器和转换[EB/OL].[202208].https://pytorch.apachecn.org/#/docs/1.4/5.
[9] Vitis AI用户指南(UG1414)2.0中文版[EB/OL].[202208].https://docs.xilinx.com/r/2.0Chinese/ug1414vitisai/VitisAI%E6%A6%82%E8%BF%B0.
基金
*广东省科技计划项目(2020A0505100015);华南理工大学智能系统未来创新实验室项目(x2jsC9226350)。
PDF(1360 KB)
