Google Releases Powerful New Gemini Robotics On-Device Model

Google DeepMind has just released a game-changing new language model called Gemini Robotics On-Device that can run complex tasks locally on robots without needing an internet connection. This latest advancement in AI and robotics technology is a major step forward for the industry.

The Gemini Robotics On-Device model builds on Google's previous Gemini Robotics model released earlier this year. This new version can directly control a robot's movements and actions using natural language prompts. Developers can fine-tune the model to suit their specific needs and applications.

In benchmarks, Google claims the Gemini Robotics On-Device model performs at a level close to the original cloud-based Gemini Robotics model. Impressively, Google says it outperforms other leading on-device AI models in general testing, though they didn't provide specifics on the competing models.

To demonstrate the capabilities of the new model, Google showed robots running the local Gemini model performing tasks like unzipping bags and folding clothes. Notably, the model was able to adapt to work on different robot hardware beyond just the ALOHA robots it was originally trained on, including the Franka FR3 bi-arm robot and the Apollo humanoid robot.

In one example, the Franka FR3 robot was able to successfully tackle new assembly tasks on an industrial belt, even for objects and scenarios it hadn't been explicitly trained on before. This highlights the model's impressive generalization abilities.

Alongside the new Gemini Robotics On-Device model, Google is also releasing a full SDK for developers to utilize. This will allow them to train robots on new tasks by showing them just 50-100 demonstrations within the MuJoCo physics simulator.

The release of Gemini Robotics On-Device is part of a broader trend of major AI companies expanding into robotics. Nvidia is building a platform for foundation models in humanoid robots, Hugging Face is developing open robotics models and datasets, and Korean startup RLWRLD is working on foundational robotics models as well.

Overall, Google's new Gemini Robotics On-Device model represents a significant breakthrough that could have wide-ranging impacts on the future of robotics. By enabling powerful AI capabilities to run locally on robot hardware, it opens up new possibilities for autonomous, intelligent systems that can operate without relying on cloud connectivity.

Developers and robotics companies should certainly keep a close eye on this technology and the Gemini Robotics SDK as they look to push the boundaries of what's possible in the field of robotics.