Google DeepMind has updated its Gemini Robotics model to manage the complete physical movements of humanoid robots, extending control from the torso down to the feet and fingers. The previous iteration handled only upper body tasks, but this new version enables actions like walking, crouching, and stretching while manipulating objects. Demonstrations feature the Apptronik Apollo 2 robot bending to pick up a watering can or removing a baseball glove from a shelf without losing balance. This shift addresses a major limitation in current robot software, which often struggles to coordinate limbs simultaneously during complex tasks. By integrating whole-body intelligence, the system allows robots to maintain stability while reaching for items or navigating uneven terrain. The update represents a practical step toward deploying humanoid machines in unstructured environments like homes or warehouses where full mobility is required.
- The model supports simultaneous lower body locomotion and upper body manipulation.
- Apptronik’s Apollo 2 was the primary testbed for these new capabilities.
- Google released the update via a blog post on Thursday.
