一句话：以因子图为统一抽象，自动为多种基于优化的机器人算法生成 FPGA 加速器。
针对基于优化的机器人应用，现有加速方案存在两类局限：一类只加速通用矩阵运算，无法充分利用机器人算法中常见的特定稀疏结构；另一类依赖手工设计专用加速器，效率低且非经常性工程（NRE）成本高昂。Orianna 框架以公共抽象——因子图（factor graph）为统一表示，克服上述问题，可为集成多种优化算法的机器人应用生成加速器。应用设计者通过 Orianna 软件框架声明目标与约束，编译器随后将高层代码翻译为基本矩阵运算，并在所生成的加速器上以乱序方式执行。作者实现了真实的 FPGA 原型进行评估：相比先进的 Intel CPU，Orianna 加速器取得 6.5 倍性能提升与 15.1 倍能耗降低，同时仅使用 36.1% 的硬件资源就达到与最先进专用加速器相当的性能。
Despite extensive efforts, existing approaches to design accelerators for optimization-based robotic applications have limitations. Some approaches focus on accelerating general matrix operations, but they fail to fully exploit the specific sparse structure commonly found in many robotic algorithms. On the other hand, certain methods require manual design of dedicated accelerators, resulting in inefficiencies and significant non-recurring engineering (NRE) costs. The Orianna framework represents a significant advancement as it overcomes the limitations of existing approaches and provides a solution to generate accelerators for robotic applications that integrate multiple optimization-based algorithms. The framework leverages a common abstraction factor graph to overcome these challenges. Application designers can easily define their objectives and constraints using the Orianna software framework. The Orianna compiler then translates the high-level code into basic matrix operations and executes them in an out-of-order manner on the generated accelerator. We demonstrate its effectiveness by implementing a real FPGA-based prototype. Orianna accelerators achieve a 6.5× performance improvement and a 15.1× energy reduction compared to an advanced Intel CPU. Furthermore, Orianna achieves comparable performance to state-of-the-art dedicated accelerators while utilizing only 36.1% of hardware resources.
@inproceedings{HaoG00HWL24,
title = {ORIANNA: An Accelerator Generation Framework for Optimization-based Robotic Applications},
author = {Yuhui Hao and Yiming Gan and Bo Yu and Qiang Liu and Yinhe Han and Zishen Wan and Shaoshan Liu},
booktitle = {ASPLOS},
year = {2024},
doi = {10.1145/3620665.3640379}
}