2026年7月30日 模型
Carolina Parada
从脚趾到指尖——我们正在教机器人智能全身控制、精细灵巧操作以及团队协作，以完成一系列广泛而复杂的任务
几十年来，我们一直梦想着机器人能够无缝融入我们的世界并伸出援手。如今，这一愿景迈出了重要一步。
大多数机器人都是针对狭窄、重复的任务序列进行预编程或远程操作的。它们缺乏真正自主学习或适应不可预测环境的能力。此外，将学到的技能从一台机器人转移到另一台机器人仍然极其困难。为了大规模解决最棘手的问题，各种形状和尺寸的机器人都需要AI模型，赋予它们思考、行动和智能交互的能力，从而安全地完成任务。
我们曾通过Gemini Robotics展示了Gemini的多模态理解如何驱动现实世界中的行动。今天，我们推出Gemini Robotics 2——为下一代真正可适应的机器人提供动力的智能层。随着它迈出实实在在的第一步，这一重大进步解锁了智能全身控制、高级灵巧操作以及多机器人协作。
Gemini Robotics 2使机器人能够对每一个动作进行推理，从而解锁广泛的任务。例如，它可以让一个人形机器人行走、下蹲、伸展并操作物体，以清理一个杂乱无章的房间。它甚至可以与其他机器人合作，更快地完成工作。这种深层次的智能还可以在设备本地运行，同时在短短几小时内无缝适应全新的机器人本体。
我们通过三个功能强大的模型实现了这一目标：
- Gemini Robotics 2：我们最先进的视觉-语言-动作模型（VLA），将视觉和语言输入转换为电机控制，使机器人能够采取行动。该模型能够控制完整的人形机器人，从脚趾到指尖，以及其他双臂机器人。它还为双手和夹爪带来了全新水平的灵巧操作能力。
- Gemini Robotics ER 2：我们最强大的具身推理（ER）模型。它是一个视觉语言模型（VLM），充当我们的智能体，使机器人能够与人类交流、理解物理世界并规划持续数分钟的多步骤任务。我们还引入了机器人作为团队协作的能力。
- Gemini Robotics On-Device 2：我们最高效的视觉-语言-动作模型（VLA），经过优化可在机器人设备上本地运行。该模型现在可以通过几小时的数据实现对新机器人本体的快速适应。
我们的推理模型Gemini Robotics ER 2现已在Google AI Studio上提供，并在Gemini Enterprise Agent Platform上提供私人预览。我们的VLA和On-Device模型可供早期访问合作伙伴使用。请在我们的开发者博客上阅读如何将这些模型应用到您的硬件上。
人形机器人在行动：管理全身任务
世界是为人类的动作而构建的；它要求我们在狭窄、杂乱的空间中伸手、弯腰和保持平衡。虽然我们之前的模型控制人形机器人的上半身来完成桌面任务，但Gemini Robotics 2将物理AI扩展到了全身动作。
我们的模型现在首次能够控制完整的人形机器人，将意图转化为智能的全身控制。例如，在控制Apptronik的Apollo 2人形机器人时，我们可以要求它_“将浇水壶放入底层货架上的绿色箱子中。”_ Apollo处理指令，走到桌子旁，拿起浇水壶，走几步到货架前，并将其精确地放置在目的地。虽然我们的机器人在移动速度方面还有待提升，但这是朝着完成需要全身协调的更复杂现实世界任务所需技能迈出的重要一步。
为双手和夹爪带来高级灵巧操作
要在我们的家庭和工作场所中真正有用，机器人需要精细的操作能力。Gemini Robotics 2在不同末端执行器上解锁了全新水平的物理灵巧性，无论是机器人使用手还是夹爪，都使机器人比以往任何时候都更有用。
该模型现在可以控制Apollo 2机器人上的五指、22自由度SharpaWave手，完成打结或密封拉链袋等精细动作。它还可以操作Franka Duo平台上的标准两指平行夹爪，执行复杂的灵巧任务（例如紧密包装）。我们正在继续提高精度和速度，以实现人类水平的灵巧性。
通过智能体推理和多机器人协作解锁高级任务
大多数现实世界任务需要在较长时间内执行多个步骤。为了管理这种复杂性，我们的具身推理（ER）模型Gemini Robotics ER 2充当机器人的高层大脑，处理用户指令并与人类交流。它观察房间，推理完成任务所需的步骤，与VLA协调执行动作，并跟踪进度直到任务完成。这种设置使机器人能够执行复杂的多步骤任务，在某个步骤失败时自我纠正，并泛化到新情况和目标。
在此次更新中，我们使机器人能够更可靠地执行更长的任务序列，持续数分钟并涉及数百个决策。Gemini Robotics ER 2现在能够理解任务何时开始和结束，并能精确定位关键事件发生的时刻，标志着进度理解能力的重大进步。
此外，我们引入了多机器人协作。这使得不同类型的机器人能够通信并协同工作，以解决单个机器人无法独自完成的复杂工作流程。
为任何机器人快速适应设备端模型
许多机器人应用需要在没有网络延迟或互联网连接的情况下运行。Gemini Robotics On-Device 2专为应对这些限制而构建——它是我们最高效的视觉-语言-动作模型（VLA），经过优化可在机器人设备上本地运行。
该模型原生支持多形态，并继承了我们在 Gemini Robotics 1.5 中先进的“动作迁移”技术。现在，我们只需几小时的适应时间，通常少于200个示例，即可适应新的双臂机器人形态。即使面对形状、传感器和自由度截然不同的新形态，这一方法同样有效，如下方由 Dexmate、SO101 和 Trossen 平台执行的各种任务所示。
推进我们对安全、负责任机器人技术的承诺
安全性是我们机器人研究的基石。随着机器人获得更多物理能力，我们致力于确保端到端的安全性与对齐。在每次发布中，我们都采取了多层方法，将传统物理安全措施与强大的 AI 安全框架相结合。
Gemini Robotics 2 特别推进了机器人安全性，以应对现实世界的不确定性并与人类协作。
我们推出了 ASIMOV-Agentic，这是一个用于智能体安全编排和不确定性解决的新基准。例如，它衡量具身推理智能体拒绝来自 VLA 的不安全工具调用的能力。它还衡量智能体预测任务是否可行，并在不确定时主动请求人工干预的能力。
此外，凭借增强的具身推理能力，Gemini Robotics ER 2 是我们迄今为止在安全约束遵循和人类接近基准方面最安全的机器人模型。它能更好地检测人类何时靠近，触发安全工具调用，并在有人过于接近时将机器人安全停止。这是协作安全标准中的关键要求。请阅读我们的 Gemini Robotics 2：安全技术报告 了解更多详情。
迈向通用物理 AI
Gemini Robotics 2 标志着在物理世界解决 AGI 道路上的重要里程碑。释放机器人技术的真正潜力需要超越单一任务自动化，迈向通用智能。通过构建这一核心智能，我们的目标是使物理世界中的 AI 能够与人类并肩工作，解决复杂挑战。
探索 Gemini Robotics 2
致谢
这项工作由 Gemini Robotics 团队开发：Abhijit Ogale、Abhishek Jindal、Adil Dostmohamed、Adrian Collister、Alan Thompson、Alessio Quaglino、Alex Bewley、Alex Hofer、Alex Taeho Kim、Alex X. Lee、Alex Zihao Zhu、Allen Chai、Amaris Paryag、Amit Hampaul、Amy Nommeots-Nomm、Amy Shen、Andre Araujo、Anirudha Majumdar、Anna Volosina、Annie S. Chen、Annie Xie、Anthony Brohan、Antoine Laurens、Arunkumar Byravan、Asaf Revach、Assaf Hurwitz Michaely、Baruch Tabanpour、Ben Moran、Benoit Landry、Bingyi Cao、Bogdan Mazoure、Brandon Hernaez、Brijen Thananjeyan、Bryan Anenberg、Caden Lu、Carl Doersch、Carolina Parada、Charles Shu、Chengda Wu、Christine Chan、Christy Koh、Chuyuan Fu、Claire Cui、Clare Lee、Claudio Fantacci、Connor Schenck、David Rendleman、Deepali Jain、Demetra Brady、Dennis Li、Dhruv Shah、Dimple Vijaykumar、Dirk Ehrlich、Divya Garikapati、Dmitry Kalashnikov、Dre Mahaarachchi、Dushyant Rao、Erik Frey、Fangchen Liu、Francesco Romano、Frankie Garcia、Gabor Simko、Gautam Salhotra、Giulia Vezzani、Grace Popple、Grace Vesom、Graziano Misuraca、Guangyao Zhou、Hagen Soltau、Hanzi Mao、Hao-Tien Lewis Chiang、Harris Chan、Hila Noga、Howard Zhou、Ian Storz、Idan Lev-Yehudi、Ignacio Rocco、Inessa Konstanz、Isaac Reid、Ishita Prasad、Ivan Kapelyukh、J. Chase Kew、Jacky Liang、Jake Varley、James Susilo、Jasmine Hsu、Jerad Kirkland、Jeremy Plassmann、Jessica Lo、Jie Tan、Jimmy Yan、Jingwei Zhang、Jinyu Xie、Jose Enrique Chen、Joshua Ainslie、Joss Moore、Juanita Bawagan、Junkyung Kim、Justin Lidard、Kanishka Rao、Kathryn Quinn Shea、Kaustubh Sridhar、Keerthana Gopalakrishnan、Ken Caluwaerts、Kenneth Oslund、Khimya Khetarpal、Konstantinos Bousmalis、Krista Reymann、Krzysztof Choromanski、Ksenia Konyushkova、Kun Zhang、Kunal Aneja、Laura Graesser、Leen Verburgh、Leonard Hasenclever、Li-Heng Lin、London Chappellet-Volpini、Lucie Kerley、Maria Attarian、Maria Bauza Villalonga、Marissa Giustina、Max McCabe、Meet Kirankumar Dave、Mehdi S. M. Sajjadi、Metin Tokosz-Exley、Michael Neunert、Michael Noseworthy、Michiel Blokzijl、Miguel Rivas、Mithun George Jacob、Mitsuhiko Nakamoto、Mo Dawoud、Mohan Kumar Srirama、Mohit Sharma、Mohit Shridhar、Muinat Abdul、Murilo F. Martins、Nathan Batchelor、Nicolas Heess、Niko Milonopoulos、Norman Di Palo、Oliver Groth、Ouais Alsharif、Padmini Copparapu、Parth Parekh、Paul Ruiz、Paul Wohlhart、Peide Huang、Peng Xu、Peter Pastor、Petko Yotov、Phil Duffy、Philemon Brakel、Rachel Sterneck、Rajkumar Vasudeva Raju、Ravin Kumar、Razvan Surdulescu、René Wagner、Reza Sanatinia、Robert Baruch、Robert Moreno、Rohan Thakker、Roland Hafner、Sajjad Zafar、Sally Jesmonth、Sam Haves、Saminda Abeyruwan、Sandy Han Huang、Scott Crowell、Seliem El-Sayed、Sergey Yaroshenko、Sergio Martinez Abad、Serkan Cabi、Sharath Maddineni、Shuang Li、Sichun Xu、Silvia Cruciani、Skanda Koppula、Skye Yang、Soo Sung、Stefan Welker、Stefani Karp、Stefano Saliceti、Steven Hansen、Stuart Bowers、Sumeet Singh、Svetlana Grant、Takahiro Miki、Takuma Yoneda、Thomas Buschmann、Thomas Lampe、Thomas Power、Thor Schaeff、Tim Hertweck、Tingnan Zhang、Todd McInally、Todor Davchev、Tong Zhao、Travers Rhodes、Tsang-Wei Edward Lee、Vika Koriakin、Vikas Sindhwani、Wenhao Yu、Wentao Yuan、Xiaolin Fang、Yahav Nussbaum、Ying Sheng、Ying Xu、Yuheng Kuang、Yuxiang Yang、Yuxiang Zhou
对于他们对这项工作的领导和支持，我们要感谢：Jean-Baptiste Alayrac、Zoubin Ghahramani、Koray Kavukcuoglu 和 Demis Hassabis。我们要感谢 Google 和 Google DeepMind 的众多团队对这项工作的贡献，包括法律、市场营销、传播、责任与安全委员会、负责任发展与创新、政策、战略与运营，以及我们的业务和企业发展团队。我们要感谢机器人团队中所有未在上述名单中明确提及的成员，感谢他们持续的支持和指导。最后，我们要感谢我们的合作伙伴：Apptronik、Boston Dynamics 和 Agile Robots 团队的支持。
July 30, 2026 Models
Carolina Parada
From feet to fingertips — we are teaching robots intelligent whole-body control, fine dexterity, and teamwork to complete a broad range of complex tasks
For decades, we’ve dreamed of robots that can seamlessly step into our world and lend a hand. Now, that vision takes a significant stride forward.
Most robots are pre-programmed or teleoperated for narrow, repetitive task sequences. They lack the ability to truly learn for themselves or adapt to unpredictable environments. Moreover, transferring learned skills from one robot body to another remains incredibly difficult. To take on the hardest problems at scale, robots of every shape and size need AI models giving them the ability to think, act, and interact intelligently to safely complete tasks.
We demonstrated how Gemini's multimodal understanding could drive real-world action with Gemini Robotics. Today, we are introducing Gemini Robotics 2 - the intelligence layer powering the next generation of truly adaptable robots. As it takes its first literal steps, this major advance unlocks intelligent whole-body control, advanced dexterity, and multi-robot collaboration.
Gemini Robotics 2 enables robots to reason through every movement, unlocking a broad range of tasks. For example, it can enable a humanoid to walk, crouch, stretch, and manipulate objects to clean up a cluttered room. It can even team up with other robots to finish the job faster. And this profound intelligence can also run locally on-device while seamlessly adapting to entirely new robotic bodies in just a few hours.
We are making this possible through three highly capable models:
- Gemini Robotics 2: Our most advanced vision-language-action model (VLA) that converts vision and language input into motor control, enabling a robot to take action. This model is capable of controlling full humanoids, from feet to fingertips, and other bi-arm robots. It also brings a new level of dexterous manipulation on both hands and grippers.
- Gemini Robotics ER 2: Our most capable embodied reasoning (ER) model. It is a vision language model (VLM) that acts as our agent, enabling robots to communicate with humans, understand the physical world and plan multi-step tasks lasting several minutes. We are also introducing the ability for robots to work together as a team.
- Gemini Robotics On-Device 2: Our most efficient vision-language-action model (VLA) optimized to run locally on robotic devices. This model can now achieve fast adaptation to completely new robot embodiments with a few hours of data.
Gemini Robotics ER 2, our reasoning model, is now available on Google AI Studio and in private preview on Gemini Enterprise Agent Platform. Our VLA and On-Device models are available to early-access partners. Read how to bring these models to your hardware on our Developer blog.
Humanoids in motion: Managing whole-body tasks
The world is built for human movements; it requires us to reach, bend, and balance in tight, cluttered spaces. While our previous models controlled the humanoid’s upper-body to achieve table-top tasks, Gemini Robotics 2 expands physical AI into whole-body motions.
For the first time, our model can now control entire humanoid robots, translating intent into intelligent whole-body control. For example, when controlling Apptronik’s Apollo 2 humanoid robot, we can ask it to “put the watering can into the green bin in the bottom shelf.” Apollo processes the instruction, walks to the table, and picks up the watering can, takes a few steps to the shelves, and places it precisely in its destination. While our robots have more to advance in movement speed, this is an important step towards the skills needed to complete more complex, real-world tasks that require whole-body coordination.
Bringing advanced dexterity to hands and grippers
To be genuinely useful in our homes and workplaces, robots need finesse. Gemini Robotics 2 unlocks a new level of physical dexterity across different end effectors, whether a robot is using hands or grippers, enabling robots to be more useful than ever before.
The model can now control the five-fingered, 22 degree-of-freedom SharpaWave hand on the Apollo 2 robot to complete delicate actions like tying knots or sealing a ziplock bag. It can also operate standard two-fingered parallel grippers on a Franka Duo platform to perform complex dexterous tasks (e.g. tight packing). We are continuing to advance the level of precision and speed to achieve human-level dexterity.
Unlocking advanced tasks with agentic reasoning and multi-robot collaboration
Most real-world tasks require multiple steps over an extended period of time. To manage this complexity, our embodied reasoning (ER) model, Gemini Robotics ER 2, serves as the robot’s high-level brain, processing user instructions and communicating with humans. It observes the room, reasons about the steps needed to complete the task, coordinates with the VLA to carry out the actions, and tracks progress until the task is done. This setup allows robots to execute complex multi-step tasks, self-correct if a step fails, and generalize to novel situations and goals.
In this update, we are enabling robots to more reliably execute longer task sequences, lasting several minutes and involving hundreds of decisions. Gemini Robotics ER 2 now understands when tasks begin and end, and can pinpoint the moment key events occur, marking a step change in progress understanding.
Furthermore, we are introducing multi-robot collaboration. This enables different types of robots to communicate and work together to solve complex workflows a single robot could not do alone.
Adapting fast on-device models for any robot
Many robotic applications need to operate without network latency or internet connectivity. Gemini Robotics On-Device 2 is built specifically to handle these constraints — it is our most-efficient vision-language-action model (VLA) optimized to run locally on robotic devices.
This model is natively multi-embodiment and inherits our advanced “motion transfer” techniques from Gemini Robotics 1.5. We can now adapt to new bi-arm robot embodiments with just a few hours of adaptation time, typically with less than 200 examples. This works even with new embodiments with drastically different shapes, sensors and degrees of freedom, as shown below with a diverse set of tasks being performed by the Dexmate, SO101, and Trossen platforms.
Advancing our commitment to safe and responsible robotics
Safety is foundational to our robotics research. As robots gain more physical capabilities, we are committed to ensuring end-to-end safety and alignment. With each release, we’ve taken a multi-layered approach that combines traditional physical safety measures with robust AI safety frameworks.
Gemini Robotics 2 specifically advances robotics safety for navigating the uncertainty of the real world and collaborating alongside humans.
We’re introducing ASIMOV-Agentic, a new benchmark for agentic safety orchestration and uncertainty resolution. For example, it measures the embodied reasoning agent’s ability to refuse unsafe tool calls from a VLA.It also measures the agent’s ability to predict whether a task is possible and to proactively request human intervention when uncertain.
Additionally, with enhanced embodied reasoning, Gemini Robotics ER 2 is our safest robotics model to date in safety constraint following and human proximity benchmarks. It can better detect when humans are nearby, trigger safety tool calls and bring the robot to a safe stop if someone approaches too closely. This is a key requirement in collaborative safety standards. Read our Gemini Robotics 2: Safety Technical Report for more details.
Building towards general-purpose physical AI
Gemini Robotics 2 marks an important milestone on the path toward solving AGI in the physical world. Unlocking the true potential of robotics requires moving past single-task automation toward general-purpose intelligence. By building this core intelligence, our goal is to enable AI in the physical world that can work alongside humans to solve complex challenges.
Explore Gemini Robotics 2
Acknowledgements
This work was developed by the Gemini Robotics team: Abhijit Ogale, Abhishek Jindal, Adil Dostmohamed, Adrian Collister, Alan Thompson, Alessio Quaglino, Alex Bewley, Alex Hofer, Alex Taeho Kim, Alex X. Lee, Alex Zihao Zhu, Allen Chai, Amaris Paryag, Amit Hampaul, Amy Nommeots-Nomm, Amy Shen, Andre Araujo, Anirudha Majumdar, Anna Volosina, Annie S. Chen, Annie Xie, Anthony Brohan, Antoine Laurens, Arunkumar Byravan, Asaf Revach, Assaf Hurwitz Michaely, Baruch Tabanpour, Ben Moran, Benoit Landry, Bingyi Cao, Bogdan Mazoure, Brandon Hernaez, Brijen Thananjeyan, Bryan Anenberg, Caden Lu, Carl Doersch, Carolina Parada, Charles Shu, Chengda Wu, Christine Chan, Christy Koh, Chuyuan Fu, Claire Cui, Clare Lee, Claudio Fantacci, Connor Schenck, David Rendleman, Deepali Jain, Demetra Brady, Dennis Li, Dhruv Shah, Dimple Vijaykumar, Dirk Ehrlich, Divya Garikapati, Dmitry Kalashnikov, Dre Mahaarachchi, Dushyant Rao, Erik Frey, Fangchen Liu, Francesco Romano, Frankie Garcia, Gabor Simko, Gautam Salhotra, Giulia Vezzani, Grace Popple, Grace Vesom, Graziano Misuraca, Guangyao Zhou, Hagen Soltau, Hanzi Mao, Hao-Tien Lewis Chiang, Harris Chan, Hila Noga, Howard Zhou, Ian Storz, Idan Lev-Yehudi, Ignacio Rocco, Inessa Konstanz, Isaac Reid, Ishita Prasad, Ivan Kapelyukh, J. Chase Kew, Jacky Liang, Jake Varley, James Susilo, Jasmine Hsu, Jerad Kirkland, Jeremy Plassmann, Jessica Lo, Jie Tan, Jimmy Yan, Jingwei Zhang, Jinyu Xie, Jose Enrique Chen, Joshua Ainslie, Joss Moore, Juanita Bawagan, Junkyung Kim, Justin Lidard, Kanishka Rao, Kathryn Quinn Shea, Kaustubh Sridhar, Keerthana Gopalakrishnan, Ken Caluwaerts, Kenneth Oslund, Khimya Khetarpal, Konstantinos Bousmalis, Krista Reymann, Krzysztof Choromanski, Ksenia Konyushkova, Kun Zhang, Kunal Aneja, Laura Graesser, Leen Verburgh, Leonard Hasenclever, Li-Heng Lin, London Chappellet-Volpini, Lucie Kerley, Maria Attarian, Maria Bauza Villalonga, Marissa Giustina, Max McCabe, Meet Kirankumar Dave, Mehdi S. M. Sajjadi, Metin Tokosz-Exley, Michael Neunert, Michael Noseworthy, Michiel Blokzijl, Miguel Rivas, Mithun George Jacob, Mitsuhiko Nakamoto, Mo Dawoud, Mohan Kumar Srirama, Mohit Sharma, Mohit Shridhar, Muinat Abdul, Murilo F. Martins, Nathan Batchelor, Nicolas Heess, Niko Milonopoulos, Norman Di Palo, Oliver Groth, Ouais Alsharif, Padmini Copparapu, Parth Parekh, Paul Ruiz, Paul Wohlhart, Peide Huang, Peng Xu, Peter Pastor, Petko Yotov, Phil Duffy, Philemon Brakel, Rachel Sterneck, Rajkumar Vasudeva Raju, Ravin Kumar, Razvan Surdulescu, René Wagner, Reza Sanatinia, Robert Baruch, Robert Moreno, Rohan Thakker, Roland Hafner, Sajjad Zafar, Sally Jesmonth, Sam Haves, Saminda Abeyruwan, Sandy Han Huang, Scott Crowell, Seliem El-Sayed, Sergey Yaroshenko, Sergio Martinez Abad, Serkan Cabi, Sharath Maddineni, Shuang Li, Sichun Xu, Silvia Cruciani, Skanda Koppula, Skye Yang, Soo Sung, Stefan Welker, Stefani Karp, Stefano Saliceti, Steven Hansen, Stuart Bowers, Sumeet Singh, Svetlana Grant, Takahiro Miki, Takuma Yoneda, Thomas Buschmann, Thomas Lampe, Thomas Power, Thor Schaeff, Tim Hertweck, Tingnan Zhang, Todd McInally, Todor Davchev, Tong Zhao, Travers Rhodes, Tsang-Wei Edward Lee, Vika Koriakin, Vikas Sindhwani, Wenhao Yu, Wentao Yuan, Xiaolin Fang, Yahav Nussbaum, Ying Sheng, Ying Xu, Yuheng Kuang, Yuxiang Yang, Yuxiang Zhou
For their leadership and support of this effort, we’d like to thank: Jean-Baptiste Alayrac, Zoubin Ghahramani, Koray Kavukcuoglu and Demis Hassabis. We’d like to recognize the many teams across Google and Google DeepMind that have contributed to this effort including Legal, Marketing, Communications, Responsibility and Safety Council, Responsible Development and Innovation, Policy, Strategy and Operations, and our Business and Corporate Development teams. We’d like to thank everyone on the Robotics team not explicitly mentioned above for their continued support and guidance. Finally, we’d like to thank our partners: Apptronik, Boston Dynamics, and Agile Robots teams for their support.
本文内容采集自官方网站，排版和翻译可能与原页面存在差异。
阅读官方全文