面向资源受限具身智能的视觉−语言−动作(VLA)模型综述

周毅; 孙煜坤; 石磊; 程玉华; 秦家虎

doi:10.16383/j.aas.c260051

面向资源受限具身智能的视觉−语言−动作(VLA)模型综述

doi: 10.16383/j.aas.c260051 cstr: 32138.14.j.aas.c260051

周毅^1,,
孙煜坤^1,,
石磊^1,,
程玉华^2,,
秦家虎^3,

1.
河南大学人工智能学院郑州 450046
2.
电子科技大学自动化工程学院成都 611731
3.
中国科学技术大学自动化系合肥 230027

基金项目: 河南省重大科技专项(251000220200, 251000210300), 河南省重点研发专项(251111211100, 261111241100) 资助

详细信息

作者简介:
周毅：河南大学人工智能学院教授. 2011年获得同济大学博士学位. 主要研究方向为认知计算, 具身智能和智能网联汽车. E-mail: zhouyi@henu.edu.cn

孙煜坤：河南大学人工智能学院硕士研究生. 主要研究方向为具身智能与多智能体系统协同. E-mail: sunyk@henu.edu.cn

石磊：河南大学人工智能学院教授. 2020年获得电子科技大学博士学位. 主要研究方向为无人系统协同定位与控制, 社交网络观点动力学. 本文通信作者. E-mail: shilei910918@126.com

程玉华：电子科技大学自动化工程学院教授. 2007年获得电子科技大学博士学位. 主要研究方向为电子测试技术, 精密无损检测, 故障诊断与健康管理. E-mail: chengyuhua_auto@uestc.edu.cn

秦家虎：中国科学技术大学自动化系教授. 2013年获得澳大利亚国立大学博士学位. 主要研究方向为多智能体系统分布式决策与控制, 信息物理系统安全与控制. E-mail: jhqin@ustc.edu.cn

计量
- 文章访问数: 5
- HTML全文浏览量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2026-01-20
- 录用日期: 2026-06-15
- 网络出版日期: 2026-07-14

Vision-Language-Action Models for Resource-constrained Embodied AI: A Survey

ZHOU Yi^1
,,
SUN Yu-Kun^1
,,
SHI Lei^1
,,
CHENG Yu-Hua^2
,,
QIN Jia-Hu^3
,

1.
School of Artificial Intelligence, Henan University, Zhengzhou 450046
2.
School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu 611731
3.
Department of Automation, University of Science and Technology of China, Hefei 230027

Funds: Supported by Science and Technology Major Project of Henan Province (251000220200, 251000210300), Key Research and Development Project of Henan Province (251111211100, 261111241100)

More Information

Author Bio:
ZHOU Yi　Professor at the School of Artificial Intelligence, Henan University. He received his Ph.D. degree from Tongji University in 2011. His research interests include cognitive computing, embodied intelligence, and intelligent connected vehicles

SUN Yu-Kun　Master student at the School of Artificial Intelligence, Henan University. His research interests include embodied intelligence and multi-agent systems coordination

SHI Lei　Professor at the School of Artificial Intelligence, Henan University. He received his Ph.D. degree from University of Electronic Science and Technology of China in 2020. His research interests include cooperative localization and control of unmanned systems, and opinion dynamics in social networks. Corresponding author of this paper

CHENG Yu-Hua　Professor at the School of Automation Engineering, University of Electronic Science and Technology of China. He received his Ph.D. degree from University of Electronic Science and Technology of China in 2007. His research interests include electronic testing technology, precision non-destructive testing, fault diagnosis and health management

QIN Jia-Hu　Professor at the Department of Automation, University of Science and Technology of China. He received his Ph.D. degree from Australian National University in 2013. His research interests include distributed decision and control of multi-agent systems, and security and control of cyber-physical systems

摘要

摘要: 视觉−语言−动作(VLA)模型是实现具身智能的重要技术路径. 针对现有方法普遍依赖大规模数据与高算力平台, 难以适应边缘侧等资源受限场景的瓶颈, 本文提出覆盖“数据−模型−训练−部署”全周期的优化框架. 在数据层面, 剖析真实采集、仿真生成与视频迁移三类路径的协同优化机制; 在设计层面, 梳理轻量化架构、状态空间模型、分层双系统、条件计算及流匹配等高效模型技术; 在系统层面, 重点探讨输入侧剪枝、模型压缩、动态推理路径及软硬件协同优化等部署加速策略. 最后, 对资源受限VLA模型的未来研究方向与应用前景进行展望.
- 具身智能 /
- 视觉-语言-动作模型 /
- 资源受限 /
- 高效模型设计 /
- 模型部署优化
Abstract: Vision-language-action (VLA) models constitute a fundamental technical pathway toward embodied intelligence. However, existing approaches are hindered by their heavy reliance on large-scale data and high-performance computing platforms, which limits deployment in resource-constrained settings such as edge devices. To address this, we propose a comprehensive optimization framework that spans the entire development cycle: Data, model, training, and deployment. At the data level, we analyze synergistic optimization mechanisms across three primary pathways: Real-world collection, simulation generation, and video transfer; At the design level, we systematically review efficient model techniques including lightweight architectures, state-space models, hierarchical dual systems, conditional computation, and flow matching; At the system level, key deployment acceleration strategies are explored, such as input pruning, model compression, dynamic inference paths, and hardware-software co-optimization. Finally, we discuss future research directions and application prospects for resource-constrained VLA models.
- embodied intelligence /
- vision-language-action models /
- resource-constrained /
- efficient model design /
- model deployment optimization

HTML全文

图 2 VLA模型训练数据的三类来源及其特性对比

Fig. 2 Three categories of training data sources for VLA models and their characteristics

下载: 全尺寸图片幻灯片

图 1 资源受限场景下VLA模型的全周期优化框架

Fig. 1 Full-lifecycle optimization framework for VLA models under resource constraints

下载: 全尺寸图片幻灯片

图 3 SmolVLA^[19]开源生态与模型架构

Fig. 3 The open-source ecosystem and model architecture of SmolVLA

下载: 全尺寸图片幻灯片

图 4 高效VLA模型的设计与部署优化技术框架

Fig. 4 Technical framework for efficient VLA model design and deployment optimization

下载: 全尺寸图片幻灯片

图 5 RoboMamba模型架构与训练流程示意图

Fig. 5 Schematic diagram of the RoboMamba model architecture and training pipeline

下载: 全尺寸图片幻灯片

图 6 三种主流VLA动作生成范式原理对比

Fig. 6 Comparison of three mainstream VLA action generation paradigms

下载: 全尺寸图片幻灯片

图 7 基于注意力热力图的视觉Token剪枝与稀疏化推理

Fig. 7 Visual Token pruning and sparse inference based on attention heatmap

下载: 全尺寸图片幻灯片

图 8 资源受限下VLA模型的未来研究展望

Fig. 8 Future research prospects for VLA models under resource constraints

下载: 全尺寸图片幻灯片

表 1 VLA模型相关综述对比

Table 1 Comparison of VLA model related surveys

综述	年份	研究主题	框架视角	主要贡献
Ma等^[10]	2024	具身智能应用	组件-策略-规划分类	建立了涵盖基础组件、低层控制与高层规划的分类体系
Zhong等^[11]	2025	动作表示机制	动作Token分类	提出VLA模型的动作Token分类框架
Zhang等^[12]	2025	VLA技术演进	历史与范式演化	回顾前VLA时代技术, 梳理三大建模范式
Wang等^[13]	2025	大模型具身智能	四级控制层级	构建需求−任务−规划−动作四级控制框架
Guan等^[8]	2025	高效VLA模型	处理流程视角	从处理流程角度系统分类高效VLA技术
Yu等^[9]	2025	高效VLA模型	数据-模型-训练	构建数据-模型-训练的高效VLA框架
本文	2025	资源受限条件	模型全生命周期	聚焦资源受限条件下的具身智能, 涵盖数据-模型-训练-部署全周期技术体系

下载: 导出CSV

表 2 面向VLA模型训练的代表性数据集总结

Table 2 Summary of representative datasets for VLA model training

分类	数据集	来源	规模	模态	主要特点
真实机器人操作数据	RT-1^[14]	真实	130 K	V、L、A	13台机械臂采集, 覆盖700余项任务
	BridgeData V2^[75]	真实	60 K	V、L、A、D	5万条遥操作, 1万条自动策略轨迹
	Open X-Emb.^[4]	真实	1 M+	V、L、A	21家机构, 22种异构机器人, 统一数据格式
	DROID^[5]	真实	76 K	V、L、A、D	564个开放场景, 分布式遥操作采集
	RH20T^[16]	真实	110 K	V、L、A、T、F、S、D	11维多模态同步, 147项接触密集型任务
	AGIBot World^[21]	真实	1 M+	V、L、A、T	217项任务, 含失败恢复轨迹与子任务标注
	RoboMIND^[76]	真实	107 K	V、L、A、D	4种平台, 479项任务, 含失败轨迹
真实-仿真混合数据	ARIO^[77]	混合	3 M+	V、L、A、T、S	258种机器人, 32万项任务, 统一数据格式
	MimicGen^[30]	混合	50 K+	V、A	200条演示自动扩展至5万条轨迹
	RoboCasa^[32]	仿真	100 K	V、A	大规模厨房场景仿真框架, 基于人工示教生成轨迹
	NVIDIA Physical AI^[78]	混合	320 K	V、L、A、D、P	基于OpenUSD的工业级仿真资产库
	RoboData^[79]	混合	500 K	V、L、A、D、C	整合9个开源数据集, 统一评测接口
人类操作视频数据	Ego4D^[80]	视频	3 670 h	V、L、S、E、I、D、P	74个地点, 3 670 h第一人称视频
	EPIC-K-100^[81]	视频	100 h	V、L、S	厨房场景, 97类动作, 9万个片段
	EgoExo4D^[82]	视频	1 286 h	V、L、E、S	第一/第三人称同步, 8个技能领域
	HOT3D^[83]	视频	13.88 h	V、E、P	多视角3D手物交互标注
	TACO^[84]	视频	2.5 K	V、D、P	2 500条双手工具操作, 高精度3D标注
	HoloAssist^[85]	视频	166 h	V、L、E、I、S	7模态同步, 人机协作场景
	UniHand^[64]	视频	150 M	V、L、A	采用动捕、VR、RGB采集, 涵盖超过1.5亿运动指令对
专项任务数据	LIBERO^[86]	仿真	6.5 K	V、L、A	程序化生成机器人操作任务, 提供遥操作示教
	REASSEMBLE^[24]	真实	4.5 K	V、L、A、F、S、C	涵盖事件相机与力觉采集, 接触密集装配任务
	RoboCerebra^[23]	仿真	1 K	V、L、A、D	平均2 972步超长轨迹, 规划能力评测
	BRMData^[87]	真实	约500	V、A、D	双臂移动平台, 10个家居场景任务
注: V =视觉, L =语言, A =动作, D =深度, T =触觉, F =力觉, S =音频, E =眼动, I = IMU, C =事件相机, P = 3D姿态/点云; 机器人数据多以轨迹/样本数计, 人类操作视频数据多以时长计, 部分数据集以序列数或样本数计.

下载: 导出CSV

表 3 主流VLA模型关键指标对比

Table 3 Comparison of key metrics for mainstream VLA models

模型	参数量	主要训练数据规模	训练开销		推理性能
模型	参数量	主要训练数据规模	训练耗时	步数/Epoch	频率(Hz)	推理平台
RT-1^[14]	35 M	13万条机器人轨迹	–	20万步	3	TPU v4
RT-2^[88]	55 B	10亿对图文与机器人数据	–	8万步	1-3	TPU v5
Octo^[89]	93 M	80万条混合轨迹	0.3万TPU h	30万步	30	RTX 4090
OpenVLA^[90]	7 B	97万条机器人轨迹	2.15万A100 h	27轮次^d	6	RTX 4090
$ \pi_0 $^[91]	3.3 B	1万小时以上机器人数据	–	70万步	$ \leq $50	RTX 4090
GR00T N1^[92]	2.2 B	5.9亿帧多源混合数据	5万H100 h	20万步^a	10/120^b	L40
GraspVLA^[40]	1.8 B	10亿帧合成抓取数据	–	12万步	5	L40s
RoboMamba^[93]	3.2 B	31万图文与机器人数据	–	5万步	9	A100
UniVLA^[42]	8.5 B	62.2万条多源视频	960 A100 h	2万余步	10	RTX 4090
SmolVLA^[19]	0.45 B	2.3万条开源轨迹	3万GPU h	20万余步	30	消费级平台
TinyVLA^[94]	1.3 B	500条微调轨迹	–	–	71^c	A6000
FLOWER^[95]	0.95 B	25万条混合轨迹	200 H100 h	36万步	311	RTX 4090
注:^a 预训练步数;^b 10 Hz(视觉语言模块)/120 Hz(动作策略模块);^c 原文指出单次推理延迟14 ms, 此处为换算得出的理论频率;^d 原文明确指出训练轮次为27 Epochs, 未披露具体训练步数. "–"表示数据未披露.

下载: 导出CSV

表 4 典型边缘计算平台硬件规格对比

Table 4 Comparison of hardware specifications of typical edge computing platforms

平台	算力	内存	功耗
NVIDIA Jetson AGX Orin	275 TOPS	64 GB	15 ~ 60 W
NVIDIA Jetson Orin NX	100 TOPS	16 GB	10 ~ 25 W
Ascend 310B	20 TOPS	24 GB	8 W
移动端NPU	约35 TOPS	共享内存	5 ~ 15 W
Raspberry Pi 5	无独立NPU	8 GB	约5 W

下载: 导出CSV

参考文献(164)

[1]	Radford A, Kim J W, Hallacy C, Ramesh A, Goh G, Agarwal S, et al. Learning transferable visual models from natural language supervision. In: Proceedings of the 38th International Conference on Machine Learning. Virtual Event: PMLR, 2021. 8748–8763
[2]	Li J, Li D, Savarese S, Hoi S C H. BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models. In: Proceedings of the 40th International Conference on Machine Learning. Honolulu, Hawaii, USA: PMLR, 2023. 19730–19742
[3]	Liu H, Li C, Wu Q, Lee Y J. Visual instruction tuning. In: Proceedings of the 37th International Conference on Neural Information Processing Systems. New Orleans, LA, USA: Curran Associates Inc., 2023. 1516–1531
[4]	O’Neill A, Rehman A, Maddukuri A, Gupta A, Padalkar A, Lee A, et al. Open X-Embodiment: Robotic learning datasets and RT-X models. In: Proceedings of IEEE International Conference on Robotics and Automation. Yokohama, Japan: IEEE, 2024. 6892–6903
[5]	Khazatsky A, Pertsch K, Nair S, Balakrishna A, Dasari S, Karamcheti S, et al. DROID: A large-scale in-the-wild robot manipulation dataset. In: Proceedings of the 20th International Conference on Robotics: Science and Systems. Delft, The Netherlands: Robotics: Science and Systems Foundation, 2024. 1–14
[6]	Oquab M, Darcet T, Moutakanni T, Vo H V, Szafraniec M, Khalidov V, et al. DINOv2: Learning robust visual features without supervision. Transactions on Machine Learning Research, 2024, 2024: 1−32
[7]	Zhai X, Mustafa B, Kolesnikov A, Beyer L. SigLIP: Sigmoid loss for language image pre-training. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. Paris, France: IEEE, 2023. 11941–11952
[8]	Guan W, Hu Q, Li A, Cheng J. Effcient vision-languageaction models for embodied manipulation: A systematic survey. arXiv: 2510.17111, 2025.
[9]	Yu Z, Wang B, Zeng P, Zhang H, Zhang J, Wang Z, et al. A survey on effcient vision-language-action models. arXiv: 2510.24795, 2025.
[10]	Ma Y, Song Z, Zhuang Y, Hao J, King I. A survey on vision-language-action models for embodied AI. IEEE Transactions on Neural Networks and Learning Systems, 2026, DOI: 10.1109/TNNLS.2025.3650584.
[11]	Zhong Y, Bai F, Cai S, Huang X, Chen Z, Zhang X, et al. A survey on vision-language-action models: An action tokenization perspective. arXiv: 2507.01925, 2025.
[12]	张慧, 梁姝彤, 李明轩, 田永林, 葛经纬, 于慧, 等. 视觉–语言–动作模型综述: 从前史到前沿. 自动化学报, 2025, 51(9): 1922−1950 doi: 10.16383/j.aas.c250417 Zhang Hui, Liang Shu-Tong, Li Ming-Xuan, Tian Yong-Lin, Ge Jing-Wei, Yu Hui, et al. Review of vision-language-action models: from prehistory to frontier. Acta Automatica Sinica, 2025, 51(9): 1922−1950 doi: 10.16383/j.aas.c250417
[13]	王文晟, 谭宁, 黄凯, 张雨浓, 郑伟诗, 孙富春. 基于大模型的具身智能系统综述. 自动化学报, 2025, 51(1): 1−19 doi: 10.16383/j.aas.c240542 Wang Wen-Sheng, Tan Ning, Huang Kai, Zhang Yu-Nong, Zheng Wei-Shi, Sun Fu-Chun. Review of large model-based embodied intelligence systems. Acta Automatica Sinica, 2025, 51(1): 1−19 doi: 10.16383/j.aas.c240542
[14]	Brohan A, Brown N, Carbajal J, Chebotar Y, Dabis J, Finn C, et al. RT-1: Robotics transformer for real-world control at scale. In: Proceedings of the 19th International Conference on Robotics: Science and Systems. Daegu, Republic of Korea: Robotics: Science and Systems Foundation, 2023. 1–16
[15]	Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A N, et al. Transformer: Attention is all you need. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, CA, USA: Curran Associates Inc., 2017. 5998–6008
[16]	Fang H, Fang H, Tang Z, Liu J, Wang C, Wang J, et al. RH20T: A comprehensive robotic dataset for learning diverse skills in one-shot. In: Proceedings of the 2024 IEEE International Conference on Robotics and Automation. Yokohama, Japan: IEEE, 2024. 653–660
[17]	Zhao T Z, Kumar V, Levine S, Finn C. Learning fine-grained bimanual manipulation with low-cost hardware. In: Proceedings of the 19th International Conference on Robotics: Science and Systems. Daegu, Republic of Korea: Robotics: Science and Systems Foundation, 2023. 1–16
[18]	Chi C, Xu Z, Pan C, Cousineau E, Burchfiel B, Feng S, et al. Universal manipulation interface: In-the-wild robot teaching without in-the-wild robots. In: Proceedings of the 20th International Conference on Robotics: Science and Systems. Delft, The Netherlands: Robotics: Science and Systems Foundation, 2024. 1–14
[19]	Shukor M, Aubakirova D, Capuano F, Kooijmans P, Palma S, Zouitine A, et al. SmolVLA: A vision-language-action model for affordable and effcient robotics. arXiv: 2506.01844, 2025.
[20]	Jiang T, Yuan T, Liu Y, Lu C, Cui J, Liu X, et al. Galaxea open-world dataset and G0 dual-system VLA model. arXiv: 2509.00576, 2025.
[21]	Bu Q, Cai J, Chen L, Cui X, Ding Y, Feng S, et al. Agi-Bot World Colosseo: Large-scale manipulation platform for scalable and intelligent embodied systems. In: Proceedings of the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems. Hangzhou, China: IEEE, 2025. 3549–3556
[22]	Jiang S, Li H, Ren R, Zhou Y, Wang Z, He B. Kaiwu: A multimodal manipulation dataset and framework for robot learning and human-robot interaction. IEEE Robotics and Automation Letters, 2025, 10(11): 11482−11489 doi: 10.1109/LRA.2025.3609615
[23]	Han S, Qiu B, Liao Y, Huang S, Gao C, Yan S, et al. RoboCerebra: A large-scale benchmark for long-horizon robotic manipulation evaluation. In: Proceedings of the 39th International Conference on Neural Information Processing Systems. San Diego, CA, USA: Curran Associates Inc., 2025. 1–14
[24]	Sliwowski D, Jadav S, Stanovcic S, Orbik J, Heidersberger J, Lee D. Demonstrating REASSEMBLE: A multimodal dataset for contact-rich robotic assembly and disassembly. In: Proceedings of the 21st International Conference on Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025. 1–14
[25]	Mildenhall B, Srinivasan P P, Tancik M, Barron J T, Ramamoorthi R, Ng R. Nerf: representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 2022, 65(1): 99−106 doi: 10.1145/3503250
[26]	Kerbl B, Kopanas G, Leimkühler T, Drettakis G. 3D gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, 2023, 42(4): 139:1−139:14 doi: 10.1145/3592433
[27]	Li X, Li J, Zhang Z, Zhang R, Jia F, Wang T, et al. RoboGSim: A real2sim2real robotic gaussian splatting simulator. arXiv: 2411.11839, 2024.
[28]	Abou-Chakra J, Rana K, Dayoub F, Sünderhauf N. Physically embodied Gaussian splatting: A visually learnt and physically grounded 3D representation for robotics. In: Proceedings of the 8th Conference on Robot Learning. Munich, Germany: PMLR, 2025. 513–530
[29]	Han X, Liu M, Chen Y, Yu J, Lyu X, Tian Y, et al. Re³Sim: Generating high-fidelity simulation data via 3D-photorealistic real-to-sim for robotic manipulation. arXiv: 2502.08645, 2025.
[30]	Mandlekar A, Nasiriany S, Wen B, Akinola I, Narang Y, Fan L, et al. MimicGen: A data generation system for scalable robot learning using human demonstrations. In: Proceedings of the 7th Conference on Robot Learning. Atlanta, GA, USA: PMLR, 2023. 1820–1864
[31]	Hua P, Liu M, Macaluso A, Lin Y, Zhang W, Xu H, et al. GenSim2: Scaling robot data generation with multi-modal and reasoning LLMs. In: Proceedings of the Conference on Robot Learning. Munich, Germany: PMLR, 2024. 5030–5066
[32]	Nasiriany S, Maddukuri A, Zhang L, Parikh A, Lo A, Joshi A, et al. RoboCasa: Large-scale simulation of household tasks for generalist robots. In: Proceedings of the 20th International Conference on Robotics: Science and Systems. Delft, The Netherlands: Robotics: Science and Systems Foundation, 2024. 1–14
[33]	Xue Z, Deng S, Chen Z, Wang Y, Yuan Z, Xu H. DemoGen: Synthetic demonstration generation for data-effcient visuomotor policy learning. In: Proceedings of the 21st International Conference on Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025. 1–14
[34]	Genesis Authors. Genesis: A generative and universal physics engine for robotics and beyond. https://github.com/Genesis-Embodied-AI/genesis-world. December 2024. Opensource software.
[35]	Mittal M, Yu C, Yu Q, Liu J, Rudin N, Hoeller D, et al. Orbit: A unified simulation framework for interactive robot learning environments. IEEE Robotics and Automation Letters, 2023, 8(6): 3740−3747 doi: 10.1109/LRA.2023.3270034
[36]	Tao S, Xiang F, Shukla A, Qin Y, Hinrichsen X, Yuan X, et al. ManiSkill3: Gpu parallelized robotics simulation and rendering for generalizable embodied AI. In: Proceedings of the 7th Robot Learning Workshop at the International Conference on Learning Representations 2025. Virtual: ICLR, 2025. 1–10
[37]	Xiang F, Qin Y, Mo K, Xia Y, Zhu H, Liu F, et al. SAPIEN: A simulated part-based interactive environment. In: Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, WA, USA: Computer Vision Foundation / IEEE, 2020. 11094–11104
[38]	Li Y, Du W, Yu C, Li P, Zhao Z, Liu T, et al. Taccel: Scaling up vision-based tactile robotics via high-performance gpu simulation. arXiv: 2504.12908, 2025.
[39]	Maddukuri A, Jiang Z, Chen L Y, Nasiriany S, Xie Y, Fang Y, et al. Sim-and-Real Co-Training: A simple recipe for vision-based robotic manipulation. In: Proceedings of Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025. 1–14
[40]	Deng S, Yan M, Wei S, Ma H, Yang Y, Chen J, et al. GraspVLA: a grasping foundation model pre-trained on billion-scale synthetic action data. arXiv: 2505.03233, 2025.
[41]	Ha D, Schmidhuber J. Recurrent world models facilitate policy evolution. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Montréal, Canada: Curran Associates, Inc., 2018. 2455–2467
[42]	Bu Q, Yang Y, Cai J, Gao S, Ren G, Yao M, et al. UniVLA: Learning to act anywhere with task-centric latent actions. In: Proceedings of Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025. 1–14
[43]	Bruce J, Dennis M D, Edwards A, Parker-Holder J, Shi Y, Hughes E, et al. Genie: Generative interactive environments. In: Proceedings of the 41st International Conference on Machine Learning. Vienna, Austria: PMLR, 2024. 5188–5221
[44]	Cen J, Yu C, Yuan H, Jiang Y, Huang S, Guo J, et al. WorldVLA: Towards autoregressive action world model. arXiv: 2506.21539, 2025.
[45]	Zhen H, Qiu X, Chen P, Yang J, Yan X, Du Y, et al. 3D-VLA: A 3D vision-language-action generative world model. In: Proceedings of the 41st International Conference on Machine Learning. Vienna, Austria: PMLR, 2024. 55141–55162
[46]	Liao Y, Zhou P, Huang S, Yang D, Chen S, Jiang Y, et al. Genie Envisioner: A unified world foundation platform for robotic manipulation. arXiv: 2508.05635, 2025.
[47]	Zhang W, Liu H, Qi Z, Wang Y, Yu X, Zhang J, et al. DreamVLA: A vision-language-action model dreamed with comprehensive world knowledge. arXiv: 2507.04447, 2025.
[48]	NVIDIA, Agarwal N, Ali A, et al. Cosmos world foundation model platform for physical AI. arXiv: 2501.03575, 2025.
[49]	Xiao J, Yang Y, Chang X, Chen R, Xiong F, Xu M, et al. World-env: Leveraging world model as a virtual environment for vla post-training. arXiv: 2509.24948, 2025.
[50]	Wu P, Escontrela A, Hafner D, Abbeel P, Goldberg K. Day-Dreamer: World models for physical robot learning. In: Proceedings of the Conference on Robot Learning. Auckland, New Zealand: PMLR, 2022. 2226–2240
[51]	Hafner D, Yan W, Lillicrap T. Training agents inside of scalable world models. arXiv: 2509.24527, 2025.
[52]	Chen D, Moutakanni T, Chung W, Bang Y, Ji Z, Bolourchi A, et al. Planning with reasoning using vision language world model. arXiv: 2509.02722, 2025.
[53]	Mendonca R, Bahl S, Pathak D. Structured world models from human videos. In: Proceedings of the 19th International Conference on Robotics: Science and Systems. Daegu, Republic of Korea: Robotics: Science and Systems Foundation, 2023. 1–16
[54]	Nair S, Rajeswaran A, Kumar V, Finn C, Gupta A. R3M: A universal visual representation for robot manipulation. In: Proceedings of the 6th Conference on Robot Learning. Auckland, New Zealand: PMLR, 2022. 892–909
[55]	Radosavovic I, Xiao T, James S, Abbeel P, Malik J, Darrell T. Real-world robot learning with masked visual pre-training. In: Proceedings of the 6th Conference on Robot Learning. Auckland, New Zealand: PMLR, 2022. 416–426
[56]	Zeng J, Bu Q, Wang B, Xia W, Chen L, Dong H, et al. Learning manipulation by predicting interaction. In: Proceedings of the 20th International Conference on Robotics: Science and Systems. Delft, The Netherlands: Robotics: Science and Systems Foundation, 2024. 1–14
[57]	Ko P-C, Mao J, Du Y, Sun S-H, Tenenbaum J B. Learning to act from actionless videos through dense correspondences. In: Proceedings of the 12th International Conference on Learning Representations. Vienna, Austria: OpenReview.net, 2024. 1–15
[58]	Oord A, Vinyals O, Kavukcuoglu K. Neural discrete representation learning. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, CA, USA: Curran Associates Inc., 2017. 6306–6315
[59]	Bahl S, Gupta A, Pathak D. Human-to-robot imitation in the wild. In: Proceedings of the 18th International Conference on Robotics: Science and Systems. New York City, NY, USA: Robotics: Science and Systems Foundation, 2022. 1–16
[60]	Allshire A, Choi H, Zhang J, McAllister D, Zhang A, Kim C M, et al. Visual imitation enables contextual humanoid control. arXiv: 2505.03729, 2025.
[61]	Patel S, Mohan S, Mai H, Jain U, Lazebnik S, Li Y. Robotic manipulation by imitating generated videos without physical demonstrations. arXiv: 2507.00990, 2025.
[62]	Yang R, Yu Q, Wu Y, Yan R, Li B, Cheng A-C, et al. EgoVLA: Learning vision-language-action models from egocentric human videos. arXiv: 2507.12440, 2025.
[63]	Romero J, Tzionas D, Black M J. Embodied hands: Modeling and capturing hands and bodies together. ACM Transactions on Graphics, 2017, 36(6): 245:1−245:17
[64]	Luo H, Feng Y, Zhang W, Zheng S, Wang Y, Yuan H, et al. Being-H0: Vision-language-action pretraining from large-scale human videos. arXiv: 2507.15597, 2025.
[65]	Cheang C-L, Chen G, Jing Y, Kong T, Li H, Li Y, et al. GR-2: A generative video-language-action model with web-scale knowledge for robot manipulation. arXiv: 2410.06158, 2024.
[66]	Jiang Y, Huang S, Xue S, Zhao Y, Cen J, Leng S, et al. RynnVLA-001: Using human demonstrations to improve robot manipulation. arXiv: 2509.15212, 2025.
[67]	Bharadhwaj H, Dwibedi D, Gupta A, Tulsiani S, Doersch C, Xiao T, et al. Gen2Act: Human video generation in novel scenarios enables generalizable robot manipulation. In: Proceedings of the 9th Conference on Robot Learning. Seoul, Republic of Korea: PMLR, 2025. 3936–3951
[68]	Xie H, Wen B, Zheng J, Chen Z, Hong F, Diao H, et al. Dynamicvla: A vision-language-action model for dynamic object manipulation. arXiv: 2601.22153, 2026.
[69]	Black K, Galliker M Y, Levine S. RTC: Real-time execution of action chunking flow policies. In: Proceedings of the 39th Annual Conference on Neural Information Processing Systems. San Diego, CA, USA: Curran Associates Inc., 2025
[70]	Sendai K, Alvarez M, Matsushima T, Matsuo Y, Iwasawa Y. Leave no observation behind: Real-time correction for VLA action chunks. arXiv: 2509.23224, 2025.
[71]	Sakr M, Zhang J, Loos H F M V d, Kulić D, Croft E. Consistency matters: Defining demonstration data quality metrics in robot learning from demonstration. Journal of Human-Robot Interaction, 2025, 15(2).
[72]	Belkhale S, Cui Y, Sadigh D. Data quality in imitation learning. In: Advances in Neural Information Processing Systems 36. New Orleans, LA, USA: Curran Associates Inc., 2023. 80375–80395
[73]	Wu K, Liu N, Zhao Z, Qiu D, Li J, Che Z, et al. Learning from imperfect demonstrations with self-supervision for robotic manipulation. In: Proceedings of the 2025 IEEE International Conference on Robotics and Automation. Atlanta, GA, USA: IEEE, 2025. 16899–16906
[74]	Hou M, Hindriks K V, Eiben A E, Baraka K. Active robot curriculum learning from online human demonstrations. In: Proceedings of the 20th ACM/IEEE International Conference on Human-Robot Interaction. Melbourne, Australia: IEEE, 2025. 810–818
[75]	Walke H R, Black K, Zhao T Z, Vuong Q, Zheng C, Hansen-Estruch P, et al. BridgeData V2: A dataset for robot learning at scale. In: Proceedings of the 7th Conference on Robot Learning. Atlanta, GA, USA: PMLR, 2023. 1723–1736
[76]	Wu K, Hou C, Liu J, Che Z, Ju X, Yang Z, et al. RoboMIND: Benchmark on multi-embodiment intelligence normative data for robot manipulation. In: Proceedings of Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025. 1–14
[77]	Wang Z, Zheng H, Nie Y, Xu W, Wang Q, Ye H, et al. All Robots in One: A new standard and unified dataset for versatile, general-purpose embodied agents. arXiv: 2408.10899, 2024.
[78]	NVIDIA Research. Physical AI: Embodied intelligence platform overview [Online], available: https://www.nvidia.com/en-us/research/physical-ai/. May 13 2025.
[79]	Yan F, Liu F, Huang Y, Guan Z, Zheng L, Zhong Y, et al. RoboTron-Mani: All-in-one multimodal large model for robotic manipulation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. Honolulu, HI, USA: IEEE, 2025. 13707–13718
[80]	Grauman K, Westbury A, Byrne E, Chavis Z, Furnari A, Girdhar R, et al. Ego4D: Around the world in 3000 hours of egocentric video. In: Proceedings of the 25th IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, LA, USA: IEEE, 2022. 18973–18990
[81]	Damen D, Doughty H, Farinella G M, Fidler S, Furnari A, Kazakos E, et al. Rescaling egocentric vision: Collection, pipeline and challenges for EPIC-KITCHENS-100. International Journal of Computer Vision, 2022, 130(1): 33−55 doi: 10.1007/s11263-021-01531-2
[82]	Grauman K, Westbury A, Torresani L, Kitani K, Malik J, Afouras T, et al. Ego-Exo4D: Understanding skilled human activity from first- and third-person perspectives. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, WA, USA: IEEE, 2024. 19383–19400
[83]	Banerjee P, Shkodrani S, Moulon P, Hampali S, Han S, Zhang F, et al. HOT3D: Hand and object tracking in 3D from egocentric multi-view videos. In: Proceedings of the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville, TN, USA: IEEE, 2025. 7061–7071
[84]	Liu Y, Yang H, Si X, Liu L, Li Z, Zhang Y, et al. TACO: Benchmarking generalizable bimanual tool-action-object understanding. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, WA, USA: IEEE, 2024. 21740–21751
[85]	Wang X, Kwon T, Rad M, Pan B, Chakraborty I, Andrist S, et al. HoloAssist: An egocentric human interaction dataset for interactive AI assistants in the real world. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. Paris, France: IEEE, 2023. 20213–20224
[86]	Liu B, Zhu Y, Gao C, Feng Y, Liu Q, Zhu Y, et al. LIBERO: benchmarking knowledge transfer for lifelong robot learning. In: Proceedings of the 37th International Conference on Neural Information Processing Systems. New Orleans, LA, USA: Curran Associates Inc., 2023. 1–15
[87]	Zhang T, Li D, Li Y, Zeng Z, Zhao L, Sun L, et al. Empowering Embodied Manipulation: A bimanual-mobile robot manipulation dataset for household tasks. arXiv: 2405.18860, 2024.
[88]	Zitkovich B, Yu T, Xu S, Xu P, Xiao T, Xia F, et al. RT-2: vision-language-action models transfer web knowledge to robotic control. In: Proceedings of the 7th Conference on Robot Learning. Atlanta, GA, USA: PMLR, 2023. 2165–2183
[89]	Ghosh D, Walke H R, Pertsch K, Black K, Mees O, Dasari S, et al. Octo: An open-source generalist robot policy. In: Proceedings of the 20th International Conference on Robotics: Science and Systems. Delft, The Netherlands: Robotics: Science and Systems Foundation, 2024. 1–16
[90]	Kim M J, Pertsch K, Karamcheti S, Xiao T, Balakrishna A, Nair S, et al. OpenVLA: An open-source vision-languageaction model. In: Proceedings of the 8th Conference on Robot Learning. Munich, Bavaria, Germany: PMLR, 2024. 2679–2713
[91]	Black K, Brown N, Driess D, Esmail A, Equi M, Finn C, et al. π₀: A vision-language-action flow model for general robot control. In: Proceedings of Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025
[92]	NVIDIA, Bjorck J, Castañeda F, Cherniadev N, Da X, et al. GR00T N1: An open foundation model for generalist humanoid robots. arXiv: 2503.14734, 2025.
[93]	Liu J, Liu M, Wang Z, An P, Li X, Zhou K, et al. RoboMamba: Effcient vision-language-action model for robotic reasoning and manipulation. In: Proceedings of the 38th International Conference on Neural Information Processing Systems. Vancouver, BC, Canada: Curran Associates Inc., 2024. 1–15
[94]	Wen J, Zhu Y, Li J, Zhu M, Tang Z, Wu K, et al. TinyVLA: Toward fast, data-effcient vision-language-action models for robotic manipulation. IEEE Robotics and Automation Letters, 2025, 10(4): 3988−3995 doi: 10.1109/LRA.2025.3544909
[95]	Reuss M, Zhou H, Rühle M, Yağmurlu O E, Otto F, Lioutikov R. FLOWER: Democratizing generalist robot policies with efficient Vision-Language-Flow models. In: Proceedings of the 9th Conference on Robot Learning. Seoul, Republic of Korea: PMLR, 2025. 3736–3761
[96]	Microsoft Research. Phi-2: The surprising power of small language models [Online], available: https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/. December 12 2023.
[97]	Zhu Y, Zhu M, Liu N, Xu Z, Peng Y. LLaVA-Phi: Effcient multi-modal assistant with small language model. In: Proceedings of the 1st International Workshop on Effcient Multimedia Computing under Limited. Melbourne, VIC, Australia: ACM, 2024. 18–22
[98]	Liang Z, Li Y, Yang T, Wu C, Mao S, Nian T, et al. Discrete diffusion VLA: Bringing discrete diffusion to action decoding in vision-language-action policies. arXiv: 2508.20072, 2025.
[99]	Wang Y, Ding P, Li L, Cui C, Ge Z, Tong X, et al. VLA-Adapter: An effective paradigm for tiny-scale vision-languageaction model. In: Proceedings of the 40th AAAI Conference on Artificial Intelligence. Singapore: AAAI Press, 2026. 18638–18646
[100]	Gu A, Dao T. Mamba: Linear-time sequence modeling with selective state spaces. In: Proceedings of the First Conference on Language Modeling. Virtual Event: OpenReview.net, 2024. 1–15
[101]	Qiao Y, Yu Z, Zhao Z, Chen S, Sun M, Guo L, et al. VL-Mamba: Exploring State Space Models for multimodal learning. In: Proceedings of the Neural Information Processing Systems Effcient Natural Language and Speech Processing Workshop 2024. Vancouver, BC, Canada: PMLR, 2024. 102–113
[102]	Li K, Li X, Wang Y, He Y, Wang Y, Wang L, et al. Video-Mamba: State Space Model for effcient video understanding. In: Proceedings of the 18th European Conference on Computer Vision. Milan, Italy: Springer, 2024. 237–255
[103]	Xie F, Nie J, Tang Y, Zhang W, Zhao H. Mamba-Adaptor: State Space Model adaptor for visual recognition. In: Proceedings of the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville, TN, USA: IEEE, 2025. 20124–20134
[104]	Park J, Park J, Xiong Z, Lee N, Cho J, Oymak S, et al. Can Mamba learn how to learn? A comparative study on in-context learning tasks. In: Proceedings of the 41st International Conference on Machine Learning. Vienna, Austria: PMLR, 2024. 39357–39379
[105]	Ye Z, Xia K, Fu Y, Dong X, Hong J, Yuan X, et al. LongMamba: Enhancing Mamba’s long-context capabilities via training-free receptive field enlargement. In: Proceedings of the 13th International Conference on Learning Representations. Singapore: OpenReview.net, 2025. 1–18
[106]	Zhang J, Guo Y, Chen X, Wang Y-J, Hu Y, Shi C, et al. HiRT: Enhancing robotic control with Hierarchical Robot Transformers. In: Proceedings of the 8th Conference on Robot Learning. Munich, Germany: PMLR, 2024. 933–946
[107]	Shentu Y, Wu P, Rajeswaran A, Abbeel P. From LLMs to actions: Latent Codes as bridges in Hierarchical Robot Control. In: Proceedings of the 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems. Abu Dhabi, United Arab Emirates: IEEE, 2024. 8539–8546
[108]	Bu Q, Li H, Chen L, Cai J, Zeng J, Cui H, et al. Towards synergistic, generalized, and effcient Dual-System for robotic manipulation. In: Proceedings of the 13th International Conference on Learning Representations. Singapore: OpenReview.net, 2025
[109]	Cui C, Ding P, Song W, Bai S, Tong X, Ge Z, et al. Open-Helix: A short survey, empirical analysis, and open-source Dual-System VLA model for robotic manipulation. arXiv: 2505.03912, 2025.
[110]	Li M, Zhao Z, Che Z, Liao F, Wu K, Xu Z, et al. SwitchVLA: Execution-aware task switching for Vision-Language-Action models. arXiv: 2506.03574, 2025.
[111]	Li Y, Meng Y, Sun Z, Ji K, Tang C, Fan J, et al. SP-VLA: A joint model scheduling and token pruning approach for vla model acceleration. arXiv: 2506.12723, 2025.
[112]	Lin F, Nai R, Hu Y, You J, Zhao J, Gao Y. OneTwoVLA: A unified vision-language-action model with adaptive reasoning. arXiv: 2505.11917, 2025.
[113]	Ho J, Jain A, Abbeel P. Denoising diffusion probabilistic models. In: Proceedings of the 34th International Conference on Neural Information Processing Systems. Virtual Event: Curran Associates Inc., 2020. 6840–6851
[114]	Lipman Y, Chen R T Q, Ben-Hamu H, Nickel M, Le M. Flow matching for generative modeling. In: Proceedings of the 11th International Conference on Learning Representations. Kigali, Rwanda: OpenReview.net, 2023. 1–20
[115]	Wang Y, Zhu H, Liu M, Yang J, Fang H-S, He T. VQ-VLA: Improving vision-language-action models via scaling vectorquantized action tokenizers. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. Honolulu, HI, USA: IEEE, 2025. 11089–11099
[116]	Pertsch K, Stachowicz K, Ichter B, Driess D, Nair S, Vuong Q, et al. FAST: Effcient action tokenization for vision-languageaction models. In: Proceedings of Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025. 1–14
[117]	Hu E J, Shen Y, Wallis P, Allen-Zhu Z, Li Y, Wang S, et al. LoRA: Low-rank adaptation of large language models. In: Proceedings of the 10th International Conference on Learning Representations. Virtual Event: OpenReview.net, 2022. 1–17
[118]	Huang D, Fang Z, Zhang T, Li Y, Zhao L, Xia C. CO-RFT: Effcient fine-tuning of vision-language-action models through chunked offline reinforcement learning. arXiv: 2508.02219, 2025.
[119]	Zhang H, Zhuang Z, Zhao H, Ding P, Lu H, Wang D. ReinboT: Amplifying robot visual-language manipulation with reinforcement learning. In: Proceedings of the 42nd International Conference on Machine Learning. Vancouver, BC, Canada: PMLR, 2025. 77254–77271
[120]	Chen Y, Tian S, Liu S, Zhou Y, Li H, Zhao D. ConRFT: A reinforced fine-tuning method for VLA models via consistency policy. In: Proceedings of Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025
[121]	Song Z, Ouyang G, Li M, Ji Y, Wang C, Xu Z, et al. ManipLVM-R1: Reinforcement learning for reasoning in embodied manipulation with large vision-language models. In: Proceedings of the 40th AAAI Conference on Artificial Intelligence. Singapore: AAAI Press, 2026. 18558–18566
[122]	Shu J, Lin Z, Wang Y. RFTF: Reinforcement fine-tuning for embodied agents with temporal feedback. arXiv: 2505.19767, 2025.
[123]	Patel S, Yin X, Huang W, Garg S, Nayyeri H, Fei-Fei L, et al. A real-to-sim-to-real approach to robotic manipulation with VLM-generated iterative keypoint rewards. In: Proceedings of the IEEE International Conference on Robotics and Automation. Atlanta, GA, USA: IEEE, 2025. 8258–8266
[124]	Zhang Z, Zheng K, Chen Z, Jang J, Li Y, Han S, et al. GRAPE: Generalizing robot policy via preference alignment. In: Proceedings of the 13th International Conference on Learning Representations. Singapore: OpenReview.net, 2025
[125]	Guo Y, Zhang J, Chen X, Ji X, Wang Y-J, Hu Y, et al. Improving vision-language-action model with online reinforcement learning. In: Proceedings of the IEEE International Conference on Robotics and Automation. Atlanta, GA, USA: IEEE, 2025. 15665–15672
[126]	Mark M S, Gao T, Sampaio G G, Srirama M K, Sharma A, Finn C, et al. Policy agnostic RL: Offline RL and online RL fine-tuning of any class and backbone. In: Proceedings of the 13th International Conference on Learning Representations. Singapore: OpenReview.net, 2025
[127]	Li H, Zuo Y, Yu J, Zhang Y, Yang Z, Zhang K, et al. SimpleVLA-RL: Scaling VLA training via reinforcement learning. arXiv: 2509.09674, 2025.
[128]	Xu S, Wang Y, Xia C, Zhu D, Huang T, Xu C. VLA-Cache: Effcient vision-language-action manipulation via adaptive token caching. In: Proceedings of the 39th Annual Conference on Neural Information Processing Systems. San Diego, CA, USA: Curran Associates, Inc., 2025
[129]	Tan X, Yang Y, Ye P, Zheng J, Bai B, Wang X, et al. Think twice, act once: Token-aware compression and action reuse for effcient inference in vision-language-action models. arXiv: 2505.21200, 2025.
[130]	Wang H, Xu J, Xiang Y, Pan J, Zhou Y, Li Y-L, et al. SpecPrune-VLA: Accelerating vision-language-action models via action-aware self-speculative pruning. arXiv: 2509.05614, 2025.
[131]	Jiang T, Jiang X, Ma Y, Wen X, Li B, Zhan K, et al. The better you learn, the smarter you prune: Towards effcient vision-language-action models via differentiable token pruning. arXiv: 2509.12594, 2025.
[132]	Duan Z, Zhang Y, Geng S, Liu G, Boedecker J, Lu C X. Fast ECoT: Effcient embodied chain-of-thought via thoughts reuse. arXiv: 2506.07639, 2025.
[133]	Chen W, Belkhale S, Mirchandani S, Pertsch K, Driess D, Mees O, et al. Training strategies for effcient embodied reasoning. In: Proceedings of the 9th Annual Conference on Robot Learning. Seoul, Republic of Korea: PMLR, 2025. 1–12
[134]	Frantar E, Ashkboos S, Hoefler T, Alistarh D. GPTQ: Accurate post-training quantization for generative pre-trained transformers. In: Proceedings of the 11th International Conference on Learning Representations. Kigali, Rwanda: OpenReview.net, 2023. 1–13
[135]	Lin J, Tang J, Tang H, Yang S, Chen W-M, Wang W-C, et al. AWQ: Activation-aware weight quantization for on-device LLM compression and acceleration. In: Proceedings of the 7th Annual Conference on Machine Learning and Systems. Santa Clara, CA, USA: mlsys.org, 2024. 87–100
[136]	Xiao G, Lin J, Seznec M, Wu H, Demouth J, Han S. SmoothQuant: Accurate and effcient post-training quantization for large language models. In: Proceedings of the 40th International Conference on Machine Learning. Honolulu, Hawaii, USA: PMLR, 2023. 38087–38099
[137]	Park S, Kim H, Jeon W, Yang J, Jeon B, Oh Y, et al. Quantization-aware imitation-learning for resource-effcient robotic control. arXiv: 2412.01034, 2024.
[138]	Ma X, Fang G, Wang X. LLM-Pruner: On the structural pruning of large language models. In: Advances in Neural Information Processing Systems 36. New Orleans, LA, USA: Curran Associates Inc., 2023. 21702–21720
[139]	Ashkboos S, Croci M L, Nascimento M G D, Hoefler T, Hensman J. SliceGPT: Compress large language models by deleting rows and columns. In: Proceedings of the 12th International Conference on Learning Representations. Vienna, Austria: OpenReview.net, 2024. 1–20
[140]	Xia M, Gao T, Zeng Z, Chen D. Sheared LLaMA: Accelerating language model pre-training via structured pruning. In: Proceedings of the 12th International Conference on Learning Representations. Vienna, Austria: OpenReview.net, 2024. 1–25
[141]	Jabbour J, Kim D-K, Smith M, Patrikar J, Ghosal R, Wang Y, et al. Don’t run with scissors: Pruning breaks VLA models but they can be recovered. arXiv: 2510.08464, 2025.
[142]	Yang Y, Wang Y, Wen Z, Zhongwei L, Zou C, Zhang Z, et al. EffcientVLA: Training-free acceleration and compression for vision-language-action models. arXiv: 2506.10100, 2025.
[143]	Chen Y, Han Y, Huang Y, Li X. RLRC: Reinforcement learning-based recovery for compressed vision-language-action models. IEEE Robotics and Automation Letters, 2026, 11: 8864−8871 doi: 10.1109/LRA.2026.3700379
[144]	Fang H, Liu Y, Du Y, Du L, Yang H. SQAP-VLA: A synergistic quantization-aware pruning framework for high-performance Vision-Language-Action models. arXiv: 2509.09090, 2025.
[145]	Wang H, Xiong C, Wang R, Chen X. BitVLA: 1-bit visionlanguage-action models for robotics manipulation. arXiv: 2506.07530, 2025.
[146]	Yue Y, Wang Y, Kang B, Han Y, Wang S, Song S, et al. DeeR-VLA: Dynamic inference of multimodal large language models for effcient robot execution. In: Proceedings of the 38th International Conference on Neural Information Processing Systems. Vancouver, BC, Canada: Curran Associates Inc., 2024. 1–15
[147]	Mees O, Hermann L, Rosete-Beas E, Burgard W. CALVIN: A benchmark for language-conditioned policy learning for longhorizon robot manipulation tasks. IEEE Robotics and Automation Letters, 2022, 7(3): 7327−7334 doi: 10.1109/LRA.2022.3180108
[148]	Zhang R, Dong M, Zhang Y, Heng L, Chi X, Dai G, et al. MoLe-VLA: Dynamic layer-skipping vision language action model via mixture-of-layers for effcient robot manipulation. In: Proceedings of the 40th AAAI Conference on Artificial Intelligence. Singapore: AAAI Press, 2026. 18764–18772
[149]	Li Q, Liang Y, Wang Z, Luo L, Chen X, Liao M, et al. CogACT: A foundational vision-language-action model for synergizing cognition and action in robotic manipulation. arXiv: 2411.19650, 2024.
[150]	Song W, Chen J, Ding P, Zhao H, Zhao W, Zhong Z, et al. PD-VLA: Accelerating Vision-Language-Action model integrated with action chunking via parallel decoding. In: Proceedings of the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems. Hangzhou, China: IEEE, 2025. 13162–13169
[151]	Song W, Chen J, Ding P, Huang Y, Zhao H, Wang D, et al. CEED-VLA: Consistency vision-language-action model with early-exit decoding. arXiv: 2506.13725, 2025.
[152]	Kim M J, Finn C, Liang P. Fine-tuning Vision-Language-Action models: Optimizing speed and success. In: Proceedings of Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025. 1–14
[153]	Wang S, Yu R, Yuan Z, Yu C, Gao F, Wang Y, et al. Spec-VLA: Speculative decoding for Vision-Language-Action models with relaxed acceptance. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. Suzhou, China: Association for Computational Linguistics, 2025. 26916–26928
[154]	Dao T, Fu D Y, Ermon S, Rudra A, R’e C. FlashAttention: Fast and memory-effcient exact attention with io-awareness. In: Proceedings of the 36th International Conference on Neural Information Processing Systems. New Orleans, LA, USA: Curran Associates Inc., 2022. 1–14
[155]	Kwon W, Li Z, Zhuang S, Sheng Y, Zheng L, Yu C H, et al. Efficient memory management for large language model serving with PagedAttention. In: Proceedings of the 29th Symposium on Operating Systems Principles. Koblenz, Germany: ACM, 2023. 611–626
[156]	Yu G-I, Jeong J S, Kim G-W, Kim S, Chun B-G. Orca: A distributed serving system for Transformer-based generative models. In: Proceedings of the 16th USENIX Symposium on Operating Systems Design and Implementation. Carlsbad, CA, USA: USENIX Association, 2022. 521–538
[157]	Xue H, Ren J, Chen W, Zhang G, Yuan F, Gu G, et al. Reactive diffusion policy: Slow-fast visual-tactile policy learning for contact-rich manipulation. In: Proceedings of Robotics: Science and Systems. Los Angeles, CA, USA: Robotics: Science and Systems Foundation, 2025. 1–14
[158]	Ma Y, Zhou Y, Yang Y, Wang T, Fan H. Running VLAs at real-time speed. arXiv: 2510.26742, 2025.
[159]	Xia C, Zhao J, Sun Q, Wang Z, Wen Y, Yu T, et al. Optimizing deep learning inference via global analysis and tensor expressions. In: Proceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems. La Jolla, CA, USA: ACM, 2024. 286–301
[160]	Shao J, Zhou X, Feng S, Hou B, Lai R, Jin H, et al. Tensor program optimization with probabilistic programs. In: Proceedings of the 36th International Conference on Neural Information Processing Systems. New Orleans, LA, USA: Curran Associates Inc., 2022. 1–14
[161]	Yang Y, Jiao L, Xu Y. A queueing theoretic perspective on low-latency LLM inference with variable token length. In: Proceedings of the 22nd International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks. Vannes, France: IEEE, 2024. 273–280
[162]	Zhen R, Li J, Ji Y, Yang Z, Liu T, Xia Q, et al. Taming the titans: A survey of effcient LLM inference serving. In: Proceedings of the 18th International Natural Language Generation Conference. Hanoi, Vietnam: Association for Computational Linguistics, 2025. 522–541
[163]	Xin J, Tang R, Lee J, Yu Y, Lin J. DeeBERT: Dynamic early exiting for accelerating BERT inference. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Online: Association for Computational Linguistics, 2020. 2246–2251
[164]	Liu J, Dai G, Xia H, Guo L, Shi X, Xu J, et al. TSTC: Two-level sparsity tensor core enabling both algorithm flexibility and hardware effciency. In: Proceedings of the 2023 IEEE/ACM International Conference on Computer Aided Design. San Francisco, CA, USA: IEEE, 2023. 1–9