Google Unveils Groundbreaking AI Model for Offline Robot Operation

In a remarkable leap forward in robotics technology, Google DeepMind has introduced a revolutionary on-device version of its Gemini Robotics AI model.

This cutting-edge vision-language-action (VLA) model is not only compact and efficient but also capable of running directly on a robot without an internet connection. This innovation opens up a world of possibilities, enabling robots to tackle a diverse array of physical tasks.

This is for even those they haven’t been specifically trained to perform. Carolina Parada, head of robotics at Google DeepMind, reveals that the original Gemini Robotics model employs a hybrid approach.

This can function seamlessly in both on-device environments and the cloud. However, with this new device-only model, users can enjoy offline capabilities that are strikingly close to those of the flagship version. 

This remarkable on-device model is equipped to perform a wide variety of tasks right out of the box and can swiftly adapt to new situations with just 50 to 100 demonstrations. Remarkably, this on-device model was trained exclusively on Google’s ALOHA robot.

However, it has demonstrated a remarkable versatility by adapting to work with various other platforms, including Apptronik’s humanoid Apollo robot and the bi-arm Franka FR3 robot. Parada expressed her excitement, stating, “While the Gemini Robotics hybrid model retains greater power, we are genuinely impressed by how effective this on-device model is.” 

In conjunction with this exciting technological launch, Google is also making available a software development kit (SDK) for developers, empowering them to evaluate and refine this groundbreaking model. 

This is an extraordinary opportunity to harness the future of robotics.