Google DeepMind has unveiled its new family of AI Gemini 2.0 models designed for multimodal robots.
To prove just how far these robots have come, DeepMind put them to the test: One robot grabbed a ball, took aim, and performed a slam dunk into a basketball hoop — something it had never done before. The feat was particularly noteworthy because, although the robots had been trained to understand natural language, the motion required sophisticated coordination.
Carolina Parada, senior director and head of robotics at Google DeepMind, wrote in a blog post that the robot completed the move on its first attempt.
This demonstration was powered by a Gemini Robotics model, which builds upon Google’s Gemini 2.0. The models have been fine-tuned with robot-specific data incorporated, and physical features have been added to Gemini’s multimodal outputs, such as text, video, and audio.
“This milestone lays the foundation for the next generation of robotics that can be helpful across a range of applications,” said Google CEO Sundar Pichai during the announcement of the new models on X.
Machines with greater dexterity
“The Gemini Robotics models are highly dexterous, interactive and general, meaning they can drive robots to react to new objects, environments, and instructions without further training,’’ according to the blog. The goal is to use AI to power robots that can assist with everyday tasks.
Google envisions robots eventually becoming another avenue for people to interact with AI, like phones and computers.
Google DeepMind is also introducing Gemini Robotics, which it touts as its most advanced vision-language-action model. Gemini Robotics aims to allow robots to comprehend something in front of them, interact with a user, and take action. It is making significant advances in dexterity, an area that has proved tricky for roboticists, Google DeepMind said.
Gemini Robotics’ machines have successfully prepared salads, packed lunches, played games such as Tic-Tac-Toe, and folded an origami fox, according to the company.
Humanoid robots are gaining momentum
Google isn’t the only company interested in seeing robots integrated into everyday life.
Companies including NVIDIA, Boston Dynamics, and Tesla are increasingly incorporating AI technology into robotic systems. By 2050, the implementation of humanoid robots is projected to grow by 61%, with more than 648 million in operation.
Meta is focusing on building AI sensors and software to power robots made and sold by other manufacturers.
NVIDIA is also delving into humanoid robotics. This year, it plans to launch its Jetson Thor platform, which aims to help robots interact more naturally with people and their surroundings.