Gemini Robotics 2 brings whole body intelligence to robots
2026-08-20 · Google DeepMind
Gemini Robotics 2 Brings Whole-Body Intelligence to Robots
On July 30, 2026, Google introduced Gemini Robotics 2, a significant advancement that delivers whole-body intelligence to robots. The system equips robots with intelligent whole-body control, fine dexterity, and the ability to collaborate in teams, enabling them to tackle a wide range of complex, real-world tasks.
For decades, robots have largely been limited to pre-programmed or teleoperated narrow, repetitive sequences. They lack independent learning capabilities, struggle to adapt to unpredictable environments, and face major challenges when transferring skills across different robot bodies. Gemini Robotics 2 addresses these limitations by acting as the intelligence layer for the next generation of truly adaptable robots across various shapes and sizes.
Three Core Models
The breakthrough is powered by three highly capable models:
Gemini Robotics 2
The most advanced vision-language-action (VLA) model. It converts vision and language inputs into motor commands, enabling full control of humanoids from feet to fingertips as well as other bi-arm robots. It delivers a new level of dexterous manipulation for both multi-fingered hands and grippers.
Gemini Robotics ER 2
The most capable embodied reasoning (ER) model. This vision-language model (VLM) serves as the robot’s high-level agent, allowing natural communication with humans, understanding of the physical world, and planning of complex multi-step tasks lasting several minutes. It also introduces the ability for robots to work together as a team.
Gemini Robotics On-Device 2
The most efficient VLA model, optimized to run locally on robotic devices. It can rapidly adapt to completely new robot embodiments using only a few hours of data.
Performance Across Different Embodiments
Using the same model checkpoint, Gemini Robotics 2 successfully controls three distinct platforms: the Apptronik Apollo 2 with SharpaWave hands, the Apollo 2 with Inspire hands, and the Franka Duo with a Robotiq gripper. The model achieves medium to high success rates on whole-body tasks and gripper-based dexterous manipulation. Multi-finger dexterous manipulation, however, remains challenging.
Humanoids in Motion: Whole-Body Control
The physical world is designed around human movement, requiring reaching, bending, and balancing in tight, cluttered spaces. While earlier models focused mainly on upper-body control for tabletop tasks, Gemini Robotics 2 extends physical AI to full whole-body coordination.
For the first time, the model can control an entire humanoid robot. In one demonstration, the Apptronik Apollo 2 is instructed to “put the watering can into the green bin in the bottom shelf.” The robot walks to the table, grasps the watering can, moves to the shelf, and places it precisely in the target location. Although movement speed still requires improvement, this marks an important milestone toward completing complex real-world tasks that demand full-body intelligence.
Advanced Dexterity for Hands and Grippers
To be genuinely useful in homes and workplaces, robots need fine motor skills. Gemini Robotics 2 unlocks significantly improved physical dexterity across different end effectors.
The model can control the 22-degree-of-freedom, five-fingered SharpaWave hand on the Apollo 2 to perform delicate actions such as tying knots or sealing a ziplock bag. It can also operate standard two-fingered parallel grippers on the Franka Duo to accomplish complex tasks like tight packing. Continued progress is being made to increase precision and speed toward human-level dexterity.
Agentic Reasoning and Multi-Robot Collaboration
Most real-world tasks involve multiple steps over extended periods. Gemini Robotics ER 2 functions as the robot’s high-level brain. It processes user instructions, observes the environment, reasons about required steps, coordinates with the VLA model to execute actions, communicates with humans, tracks progress, and self-corrects when necessary.
This architecture enables robots to handle complex, long-horizon tasks, recover from failures, generalize to novel situations and goals, and collaborate efficiently with other robots to complete jobs faster.
Gemini Robotics ER 2 is now available on Google AI Studio and in private preview on the Gemini Enterprise Agent Platform. The VLA and On-Device models are available to early-access partners. Guidance on integrating these models with hardware is provided on the Developer blog.
(612 words)