Google Unveils AI Model for Robotics, Competing with Meta and OpenAI
Google’s Latest Advances in Robotics with AI Models
Alphabet’s artificial intelligence lab is launching two innovative models aimed at improving robotics. These models are designed to assist developers in training robots to react to unfamiliar situations—an ongoing challenge in robot technology.
Gemini Robotics and Gemini Robotics-ER
Google’s research division, Google DeepMind, has introduced Gemini Robotics, which focuses on creating robots that can be more skillful and interactive. Additionally, the Gemini Robotics-ER model is tailored for spatial awareness, enabling robot manufacturers to leverage Gemini’s reasoning capabilities in their new projects.
DeepMind engineer Kanishka Rao expressed that the application of Gemini in robotics brings the possibility of developing "general purpose robotics." He pointed out that, to operate efficiently in our complex and ever-changing environments, robots need to manage diverse and unpredictable tasks.
A Surge of Interest in Robotics
The aspiration to construct robots that perform equivalent tasks to humans is gaining significant momentum, attracting fresh investments and attention from notable tech companies. Major players like Meta, Tesla, and OpenAI are intensifying their focus on robotics, leading to startups seeking funding opportunities evaluated at high valuations.
Demonstrating Robotic Capabilities
In a recent demonstration, Google showcased robots utilizing their new technology responding to straightforward commands. One notable instance involved a robot in front of a set of letter tiles, successfully spelling “Ace” upon request.
The researchers also set up a small basketball court and instructed another robot to perform a dunk. Impressively, the robot executed the task by pushing a small plastic ball through the hoop—a feat made even more remarkable as the robot had no prior exposure to basketball. This accomplishment exemplified the robot’s ability to grasp general concepts and effectively apply them in real-world settings, as highlighted by Rao.
Google’s Robotics Journey
Google’s history in the robotics sector has seen its share of challenges. Over a decade ago, the company acquired at least eight robotics firms to align with co-founders Larry Page and Sergey Brin’s vision of creating consumer-focused robots using machine learning. As efforts progressed, they became centralized within Google X, Alphabet’s innovation lab. In 2021, a subsidiary named Everyday Robots was formed to develop robots capable of everyday tasks, such as sorting waste. However, this unit was shut down in 2023 as part of cost-cutting measures.
Despite the setbacks, Alphabet has remained engaged in robotics by integrating technology and personnel into existing initiatives. Now, the company is revitalizing its robotics pursuit under the umbrella of generative AI.
Focus on Safety in Robotics
During a recent media briefing, Google emphasized the early-stage nature of this work. According to Vikas Sindhwani, a research scientist at DeepMind, the Gemini models were created with a solid foundation in “common sense safety” considerations for physical environments. The plan involves gradually deploying the robots, starting at safe distances from human interaction and aiming for increased collaboration as safety protocols become more robust.
Collaborations with Industry Leaders
Google is set to explore the capabilities of Gemini Robotics in collaboration with companies actively involved in robotics. Apptronik, known for developing humanoid robots, is one of the partners in this venture. Furthermore, Agile Robots and Boston Dynamics—previously acquired and then sold to SoftBank Group—are also involved in testing the Gemini Robotics-ER model.
By fostering these partnerships, Google aims to advance the field of robotics significantly, enhancing the functionality and safety of robots in diverse environments.
