•  
  •  
 

Bulletin of Chinese Academy of Sciences (Chinese Version)

Keywords

embodied AI；dataflow computing；chiplet；memory-centric architecture；RISC-V；benchmarking and standards

Abstract

Embodied AI is emerging as a key paradigm empowering general-purpose autonomy, but it requires on-device computing systems that simultaneously support high-throughput “cognition–planning” tasks and millisecond-level real-time “perception–control” loops. Converging solutions now coalesce around three pillars: (1) dataflow- and chiplet-based architectures, (2) memory-centric heterogeneous dies, and (3) RISC-V customizable cores with open tool-chains. This study distills the latest technical progress, pinpoints the remaining core technical bottlenecks, and charts an actionable course for academia, industry, and policymakers. The study calls for unified benchmarking and standardization, open-source software–hardware ecosystems, memory-centric dataflow architectures, and efficient on-device deployment of embodied foundation models. Finally, it outlines a layered roadmap and application scenarios to transform technological breakthroughs into industrial impact and international rule-making capacity.

First page

1936

Last Page

1947

Language

Chinese

Publisher

Bulletin of Chinese Academy of Sciences

References

[1] Brooks R A. Intelligence without representation. Artificial Intelligence, 1991, 47(1-3): 139-159.

[2] Wilson M. Six views of embodied cognition. Psychonomic Bulletin & Review, 2002, 9(4): 625-636.

[3] Liu S S. Shaping the outlook for the autonomy economy. Communications of the ACM, 2024, 67(6): 10-12.

[4] Wu S, Yu B, Liu S S, et al. Autonomy 2.0: The quest for economies of scale. Communications of the ACM, 2025, 68(4): 28-32.

[5] Radford A, Kim J W, Hallacy C, et al. Learning transferable visual models from natural language supervision// Proceedings of the 38th International Conference on Machine Learning. Online: PMLR, 2021: 8748-8763.

[6] Dong Q F, Chen X L, Satyanarayanan M. Creating edge AI from cloud-based LLMs// Proceedings of the 25th International Workshop on Mobile Computing Systems and Applications. San Diego: ACM, 2024: 8-13.

[7] Vaswani A. Attention is all you need// Advances in Neural Information Processing Systems. Red Hook: Curran Associates, 2017: 5998-6008.

[8] Liu S S, Liu L K, Tang J, et al. Edge computing for autonomous driving: Opportunities and challenges// Proceedings of the IEEE. Piscataway: IEEE, 2019: 1697-1716.

[9] Jouppi N P, Hyun Yoon D, Ashcraft M, et al. Ten lessons from three generations shaped google's TPUv4i: Industrial product// 2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (ISCA). Valencia: IEEE, 2021: 1-14.

[10] 程学旗, 梅宏, 赵伟, 等. 数据科学与计算智能：内涵、范式与机遇. 中国科学院院刊, 2020, 35(12): 1470-1481. Cheng X Q, Mei H, Zhao W, et al. Data science and computing intelligence: Concept, paradigm, and opportunities. Bulletin of Chinese Academy of Sciences, 2020, 35(12): 1470-1481. (in Chinese)

[11] Liu A S, Zhou Y G, Liu X L, et al. Compromising embodied agents with contextual backdoor attacks. (2024-08-06). https://arxiv.org/abs/2408.02882.

[12] Brohan A, Brown N, Carbajal J, et al. RT-2: Vision-language-action models transfer web knowledge to robotic control. (2023-07-28). https://arxiv.org/abs/2307.15818.

[13] Driess D, Xia F, Sajjadi M S M, et al. PaLM-E: An embodied multimodal language model. (2023-03-06). https://arxiv.org/abs/2303.03378.

[14] Liu S, Gaudiot J L. IEEE International Roadmap for Devices and Systems (IRDS) autonomous machine computing white paper 2024. Piscataway: IEEE, 2024.

[15] Huang Y Y, Hao Y H, Yu B, et al. Dadu-corki: Algorithm-architecture co-design for embodied AI-powered robotic manipulation// Proceedings of the 52nd Annual International Symposium on Computer Architecture. Tokyo: ACM, 2025: 327-343.

[16] 刘少山, 甘一鸣, 韩银和. 机器人计算发展研究. 中国科学院院刊, 2024, 39(11): 1956-1965. Liu S S, Gan Y M, Han Y H. Overview on autonomous machine computing. Bulletin of Chinese Academy of Sciences, 2024, 39(11): 1956-1965. (in Chinese)

[17] Chen Y H, Krishna T, Emer J S, et al. Eyeriss: An energy-efficient reconfigurable accelerator for deep convolutional neural networks. IEEE Journal of Solid-State Circuits, 2017, 52(1): 127-138.

[18] Jouppi N P, Young C, Patil N, et al. In-datacenter performance analysis of a tensor processing unit// Proceedings of the 44th Annual International Symposium on Computer Architecture. Toront: ACM, 2017: 1-12.

[19] Yu B, Hu W, Xu L M, et al. Building the computing system for autonomous micromobility vehicles: Design constraints and architectural optimizations// 2020 53rd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO). Athens: IEEE, 2020: 1067-1081.

[20] 包云岗, 孙凝晖. 开源芯片生态技术体系构建面临的机遇与挑战. 中国科学院院刊, 2022, 37(1): 24-29. Bao Y G, Sun N H. Opportunities and challenges of building CPU ecosystem with open-source mode. Bulletin of Chinese Academy of Sciences, 2022, 37(1): 24-29. (in Chinese)

[21] Chi P, Li S C, Xu C, et al. PRIME: A novel processing-in-memory architecture for neural network computation in ReRAM-based main memory// 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA). Seoul: IEEE, 2016: 27-39.

[22] Chih Y D, Lee P H, Fujiwara H, et al. 16.4 an 89TOPS/W and 16.3TOPS/mm2 all-digital SRAM-based full-precision compute-In memory macro in 22nm for machine-learning edge applications// 2021 IEEE International Solid-State Circuits Conference (ISSCC). San Francisco: IEEE, 2021: 252-254.

[23] Ke L, Gupta U, Cho B Y, et al. RecNMP: Accelerating personalized recommendation with near-memory processing// 2020 ACM/IEEE 47th Annual International Symposium on Computer Architecture (ISCA). Valencia: IEEE, 2020: 790-803.

[24] 高文. 抢抓人工智能发展的历史性机遇——深刻领会习近平总书记关于人工智能的重要论述. 中国信息化, 2025, (2): 6-9. Gao W. Seizing the historic opportunity for the development of artificial intelligence: Deeply comprehend general secretary Xi Jinping's important expositions on artificial intelligence. China Information Technology, 2025, (2): 6-9. (in Chinese)

Download
Request a Copy

Share

COinS