Gemini Robotics 2 brings whole-body intelligence to robots
Next-gen robots can now move like humans and perform complex tasks.
1Whole-body control
Gemini Robotics 2 enables human-like whole-body movement capabilities in robots. The Apollo 2 humanoid can now understand and execute complex commands like 'put the green box on the bottom shelf.' It coordinates walking, bending, reaching, and object manipulation. This capability marks a critical breakthrough for real-world tasks.
2Advanced dexterity
The system can perform fine motor skills like tying knots and sealing ziplock bags with its 22-degree-of-freedom SharpaWave hand. It also handles complex tasks like tight packing with standard two-finger grippers. Google continues to improve precision and speed to achieve human-level dexterity.
3Multi-robot collaboration
The Gemini Robotics ER 2 model provides high-level reasoning capabilities. Robots can now manage task sequences lasting minutes and requiring hundreds of decisions. Different robot types can communicate to solve complex workflows impossible for single robots. This feature has revolutionary potential for industrial applications.
4Rapid adaptation
Gemini Robotics On-Device 2 is optimized to work without network connectivity. It can adapt to new robot embodiments in just hours. Motion transfer technology allows learned skills to transfer across different robot platforms. This capability offers significant flexibility for field applications.
5Success rates
Gemini Robotics 2 achieves medium-high success rates in whole-body and gripper-based tasks. However, multi-finger manipulation remains challenging. Tested across platforms like Apptronik Apollo 2 and Franka Duo, the system performs well across a wide task spectrum.
6Availability
Gemini Robotics ER 2 is now available on Google AI Studio and in private preview on Gemini Enterprise Agent Platform. VLA and On-Device models are open to early-access partners. Developers can access integration guides on the Developer blog.
7Industrial impact
This technology has broad applications from production lines to logistics. It could replace human workers in hazardous environments or precision tasks. By 2026, 35% of the global industrial robot market is expected to adopt such intelligent systems.
8Technical architecture
The system comprises three core models: VLA (Vision-Language-Action) handles real-time perception and action. ER (Embodied Reasoning) manages high-level planning. On-Device provides a lightweight version for offline operation. This architecture enables autonomous operation in complex environments.
9Future vision
Google aims for full integration of robots into human environments. By 2030, robots working alongside humans in homes, hospitals, and offices are expected to become common. This development will likely cause fundamental shifts in labor markets worldwide.
Original article: DeepMind — Gemini Robotics 2 ile robotlara bütünsel zeka