In SCI-Fi stories, artificial intelligence often feeds all kinds of smart, capable and occasional murderous robots. A revealing limitation of the best AI of today is that the square is currently imprisoned in the chat window.
Google DeepMind has indicated a plan to change that today – presumably minus the murder – by announcing a new version of his AI model that merges language, vision and physical action to drive a series of more capable, adaptive and potentially useful robots.
In a series of demonstration videos, the company showed various robots that are equipped with the new model, called Gemini Robotics, manipulate items in response to spoken commands: robot arms folding paper, hand over vegetables, carefully put a glasses in a housing and complete other tasks. The robots rely on the new model to connect items that are visible with possible actions to do what they have been told. The model is trained in a way so that behavior can be generalized over very different hardware.
Google DeepMind also announced a version of his model called Gemini Robotics-er (for embodied reasoning), which only has visual and spatial understanding. The idea is that other robot researchers use this model to train their own models to control the actions of robots.
In a video demonstration, the Google DeepMind researchers used the model to control a humanoid robot called Apollo, from the Startup Apptronik. The robot talks to a person and moves letters around a table top when it is instructed.
“We were able to bring the worldwide concept of general concept understanding of Gemini 2.0 to robotics,” said Kanishka Rao, a robotics researcher at Google Deepmind who led the work, on a briefing for today's announcement.
Google DeepMind says that the new model is able to successfully manage different robots in hundreds of specific scenarios that have not been included in their training before. “As soon as the robot model has a general concept, it becomes much more general and more useful,” said Rao.
The breakthroughs that gave rise to powerful chatbots, including the chatgpt of OpenAi and Google's Gemini, have increased the hope of a similar revolution in robotics in recent years, but there will continue to be major obstacles.
