Google DeepMind has released a new artificial intelligence model, Gemini Robotics On-Device, which can perform tasks locally on robots without an Internet connection and control their movements.
Google DeepMind has released a new artificial intelligence model, Gemini Robotics On-Device, which can perform tasks locally on robots without an Internet connection and control their movements.
As TechCrunch notes, Gemini Robotics On-Device allows developers to control and customize the model to suit different needs using natural language prompts. Google claims that the model performs at a level close to Gemini Robotics’ cloud-based model.
In a demo, the company showed robots running on this local model unpacking bags and folding clothes. Google says that while the model was trained on ALOHA robots, it was later adapted to work on the Franka FR3 bi-armed robot and Apptronik’s Apollo humanoid robot. It is noted that the Franka FR3 bi-armed robot successfully coped with scenarios and objects that it had not previously «seen», such as assembly on an industrial assembly line.
Google DeepMind is also releasing the Gemini Robotics SDK. The company said developers can show robots 50 to 100 task demonstrations to train them on new tasks using these models on the MuJoCo physical simulator.
The team sees Gemini Robotics On-Device as an ideal solution for environments where cloud connectivity is unstable or non-existent. Local processing of the robot’s visual data is also better for privacy, such as in healthcare.
Figure AI has previously released an update to its Helix artificial intelligence system, which enables humanoid robots to perform complex tasks. This update improves their ability to work in sorting centers and handle packages.
