Google Introduces Gemini Robotics On-Device AI Model for Robots

Google has introduced Gemini Robotics On-Device, a new AI language model for robots. This model allows robots to operate offline and can be fine-tuned using natural language prompts. Initial tests show it performs comparably to its cloud counterpart and excels in various benchmarks. The model is designed for use with ALOHA robots, Franka FR3, and Apptronik's Apollo humanoid robot. Google also launched the Gemini Robotics SDK to support developers in training robots in new tasks.

Google DeepMind has announced the launch of Gemini Robotics On-Device, a groundbreaking AI language model designed for local operation on robotic devices. This new model, which builds upon the original Gemini Robotics, is optimized to run efficiently on the robot itself, eliminating the need for internet connectivity [1].

The Gemini Robotics On-Device model demonstrates strong general-purpose dexterity and rapid task adaptation, making it suitable for a variety of applications. It can handle tasks such as unzipping bags, folding clothes, and executing complex industrial tasks, all while operating directly on the robot [1].

One of the key advantages of Gemini Robotics On-Device is its ability to generalize across different robotic embodiments. Initial tests have shown that the model can be adapted to work with ALOHA robots, Franka FR3, and Apptronik's Apollo humanoid robot, indicating its versatility and adaptability [1].

Google DeepMind has also introduced the Gemini Robotics SDK, a toolkit designed to assist developers in fine-tuning the model for new tasks. The SDK allows for quick adaptation with as few as 50 to 100 demonstrations, making it easier for developers to integrate the model into their projects [1].

The introduction of Gemini Robotics On-Device marks a significant step forward in the field of robotics. By providing a model that can operate independently of a data network, it addresses important latency and connectivity challenges, making it more accessible and adaptable for a wide range of applications [1].

References:
[1] https://deepmind.google/discover/blog/gemini-robotics-on-device-brings-ai-to-local-robotic-devices/
[2] https://www.theinformation.com/briefings/softbank-proposes-1-trillion-facility-ai-robotics
[3] https://www.theverge.com/news/691882/google-deepmind-on-device-ai-robots-gemini