Google DeepMind launched the latest language model known as Gemini Robotics On-Device that efficiently runs tasks locally on robots offline.
Developed on the platform’s previous Gemini Robotics model that was launched in March, Gemini Robotics On-Device is able to control robot’s movements.
Developers can easily control and refine the model to meet the requirements of using natural language prompts.
The Alphabet-owned firm claimed that the model has outperformed several other models at a level close to the cloud-based Gemini Robotics model.
In a demo, Google displayed robots running this local model performing several tasks such as unzipping bags and folding clothes.
The company further claimed that the bi-arm Franka FR3 successfully tackled scenarios and objects it hadn’t “seen” before, like doing assembly on an industrial belt.
It is important to note that Google DeepMind is also launching a Gemini Robotics SDK.
The company stated that developers will be able to display robots 50 to 100 demonstrations of tasks to train them on different tasks using these models on the MuJoCo physics simulator.
In addition, Nvidia is developing a platform to produce foundation models for humanoids; Hugging Face is not only developing open models and datasets for robotics, it is currently building on robots too.