Google DeepMind has unveiled Gemini Robotics 2, an updated vision-language-action model featuring intelligent whole-body control, advanced dexterity, and multi-robot collaboration. According to Google DeepMind, the new system enables robots to reason through every movement to handle complex, multi-step physical tasks in real-world environments.
Whole-Body Control and Advanced Dexterity
Previous iterations largely focused on upper-body tabletop manipulation. Gemini Robotics 2 expands physical artificial intelligence into whole-body motions, allowing humanoids to walk, crouch, stretch, and manipulate objects simultaneously. When paired with Apptronik’s Apollo 2 humanoid robot, users can issue commands such as placing a watering can on a specific shelf, prompting the hardware to walk across the room, pick up the item, and execute the delivery precisely. To achieve finesse, the model controls end effectors ranging from standard two-fingered parallel grippers on a Franka Duo platform to the five-fingered, 22 degree-of-freedom SharpaWave hand on Apollo 2 for delicate actions like tying knots.
Multi-Robot Collaboration and Embodied Reasoning
To manage workflows lasting several minutes and involving hundreds of decisions, Google DeepMind introduced Gemini Robotics ER 2. This embodied reasoning model acts as the high-level brain, processing user instructions, observing surroundings, and communicating with humans. According to the company, it coordinates directly with the vision-language-action model, tracks progress, and self-corrects if a step fails. Furthermore, the system enables multi-robot collaboration, allowing different types of hardware to communicate and work together on tasks a single machine cannot complete alone.
Local Execution and Rapid Embodiment Adaptation
Many deployment environments suffer from network latency or lack internet connectivity entirely. To address these operational constraints, Google DeepMind optimized Gemini Robotics On-Device 2 to run locally on robotic hardware. The model inherits motion transfer techniques from earlier versions, allowing it to adapt to completely new robotic bodies—including bi-arm embodiments with drastically different shapes, sensors, and degrees of freedom—in just a few hours using typically fewer than 200 examples, according to Google DeepMind.
Safety Benchmarks and Availability
Google DeepMind has introduced ASIMOV-Agentic, a new benchmark designed to measure agentic safety orchestration and uncertainty resolution, such as an embodied reasoning agent’s ability to refuse unsafe tool calls or proactively request human intervention. Gemini Robotics ER 2 is available now on Google AI Studio and in private preview on the Gemini Enterprise Agent Platform, while the vision-language-action and on-device models are accessible to early-access partners.
Did You Know?
Gemini Robotics On-Device 2 can adapt to entirely new robotic configurations using fewer than 200 training examples, cutting down setup time to just a few hours.
Frequently Asked Questions
What is Gemini Robotics 2?
It is a vision-language-action model by Google DeepMind that provides intelligent whole-body control, advanced dexterity, and multi-robot collaboration for robotic systems.
How does Gemini Robotics ER 2 work?
It serves as a high-level embodied reasoning brain that processes human instructions, plans multi-step tasks lasting several minutes, and coordinates with local models to execute actions.
Can Gemini Robotics On-Device 2 run without internet connectivity?
Yes, the On-Device 2 model is optimized to run locally on robotic devices to eliminate network latency and operate independently of internet access.
Where are these models currently available?
Gemini Robotics ER 2 is available on Google AI Studio and in private preview on the Gemini Enterprise Agent Platform, while the vision-language-action and on-device models are available to early-access partners.
Want to stay updated on the latest breakthroughs in robotics and physical AI? Subscribe to our newsletter or explore our site for more in-depth coverage.
Keep reading