| 耿国帅* **,李早霞***.HyEA-Gen:超弹材料柔性结构计算加速器生成框架[J].高技术通讯(中文),2026,36(8):823~835 |
| HyEA-Gen:超弹材料柔性结构计算加速器生成框架 |
| HyEA-Gen: generating accelerators for hyperelastic structure computing |
| |
| DOI：10. 3772 / j. issn. 1002 - 0470. 2026. 08. 005 |
| 中文关键词: 刚柔机器人； 本构模型； 加速器生成框架； 领域知识优化 |
| 英文关键词: hybrid rigid-flexible robots, constitutive model, accelerator generation framework, domainknowledge based optimization |
| 基金项目: |
| 作者 | 单位 | | 耿国帅* ** | (*中国科学院计算技术研究所智能计算机研究中心北京 100190)
(**中国科学院大学北京 100049)
(***安阳师范学院物理与电气工程学院安阳 455000) | | 李早霞*** | |
|
| 摘要点击次数: 20 |
| 全文下载次数: 20 |
| 中文摘要: |
| 刚性和柔性结构的混合使用是机器人设计中的一个重要趋势，而柔性结构模型的计算复杂度带来的控制实时性问题限制了刚柔混合机器人的发展。本工作针对刚柔混合机器人中柔性结构模型的计算加速需求，设计了专用加速器自动生成框架HyEA-Gen，兼顾了计算模型灵活性和计算性能。该框架由两部分构成：其一为高层次综合器，用于从高层次语言（如Matlab，C++等）生成硬件加速器设计；其二为领域知识优化器，用于利用领域知识对电路结构进行优化，加速性能。通过覆盖广泛应用领域的多种柔性材料模型的实验测试，验证了设计方法的正确性和先进性。相比通用中央处理器(central processing unit，CPU)平台和Vitis HLS(high-level synthesis)生成的现场可编程门阵列(field-programmable gate array，FPGA)实现，HyEA-Gen分别实现了208.8×和14.1×的加速比。 |
| 英文摘要: |
| The use of flexible structures in rigid robot is an important trend in robot design. However, the computational complexity of flexible structure poses challenges to real-time control, limiting the development of hybrid rigid-flexible robots. This work addresses this problem by proposed HyEA-Gen, a framework for dedicated accelerator generation, which balances the flexibility and performance of computational models. The framework consists of two parts: (1)A high-level synthesizer, which generates accelerator from high-level languages (such as Matlab, C++, etc.); (2)A domain optimizer, which leverages domain-specific knowledge to optimize circuit and enhance acceleration performance. Through experiments covering a variety of flexible material models widely used in applications, the correctness and advancement of HyEA-Gen have been verified. Compared with general-purpose central processing unit central processing unit (CPU) platforms and field-programmable gate array (FPGA) implementations generated by Vitis HLS (high-level synthesis), HyEA-Gen achieves speedup ratios of 208.8× and 14.1×, respectively. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
