Simultaneous localization and mapping (SLAM) becomes a solved problem in terms of the algorithms. In order to use them in autonomous mobile robots, researches are focusing on reducing computation complexity and developing embedded systems with appropriate software and hardware architecture. In this work, we aim to perform a fully embedded implementation of a SLAM on an FPGA based SoC combining an embedded Hard Processor System (HPS) with an FPGA in the same chip. The followed implementation methodology is based on software profiling. The SLAM is profiled to perform an appropriate hardware/software partitioning and decide which part of the SLAM will be accelerated on the FPGA. The software is running on Linux on the HPS of the SoC-FPGA and the accelerated function is executed on the FPGA. The FPGA is programmed using OpenCL which is a high level language that can reduce significantly the developpement time. The results show that the SLAM algorithm implemented on SoC-FPGA is accelerated to nearly the twice comparing to the pure software implementation, while ensuring the real-time constraint

就算法而言，同时定位和映射（SLAM）已成为一个已解决的问题。为了在自动移动机器人中使用它们，研究集中在降低计算复杂性和开发具有适当软件和硬件体系结构的嵌入式系统上。在这项工作中，我们旨在在基于FPGA的SoC上实现SLAM的完全嵌入式实现，在同一芯片中将嵌入式硬核处理器系统（HPS）与FPGA结合在一起。遵循的实现方法基于软件配置文件。对SLAM进行了分析，以执行适当的硬件/软件分区，并确定将在FPGA上加速SLAM的哪一部分。该软件在SoC-FPGA的HPS上的Linux上运行，加速功能在FPGA上执行。FPGA使用OpenCL编程，OpenCL是一种高级语言，可以显着减少开发时间。结果表明，在确保实时约束的同时，在SoC-FPGA上实现的SLAM算法比纯软件实现快了近两倍。