Google DeepMind’s Gemini Robotics 2 Advances Full-Body Control for Humanoids
The latest Gemini Robotics 2 model enables humanoid machines to synchronize movements across their entire frames. This upgrade expands beyond simple arm activities to include walking, crouching, and complex object retrieval.
Key takeaways
- Gemini Robotics 2 allows AI to control a humanoid robot's entire body rather than just its arms.
- Integrated movements now include walking, crouching, and stretching alongside object manipulation.
- Demonstrations featured Apptronik’s Apollo 2 robot performing complex tasks like picking up a watering can.
- The official announcement for this model update was made by Google DeepMind on Thursday.
Comprehensive Motion Integration
Google DeepMind announced a significant technical shift on Thursday with the introduction of Gemini Robotics 2. While earlier iterations focused primarily on upper-body dexterity, this new AI model provides synchronized control across a robot’s entire architecture. The framework now manages everything from foot positioning to fingertip precision, allowing for a more fluid range of physical orientations.
Expanding Physical Capabilities
By integrating whole-body mechanics, the model enables humanoids to execute a variety of foundational movements that were previously segmented. These include:
- Crouching and Stretching: Low-profile maneuvers and extended reaching patterns.
- Locomotion: Coordinated walking while maintaining stability.
- Object Manipulation: Finding and retrieving specific items from varied Heights.
Visual demonstrations released by the company highlight the software in action using Apptronik’s Apollo 2 hardware. In one instance, the robot identifies and retrieves a baseball glove from a storage shelf; in another, it bends down to grasp a watering can. These demonstrations signify a move toward many-task versatility where machines can navigate and interact with environments in a more human-like manner.
Source: The Verge