Google DeepMind unveiled groundbreaking advancements in robotic intelligence through their Gemini AI technology. According to a newly published research paper, the DeepMind team is leveraging Gemini 1.5 Pro to enhance their robots’ ability to navigate and complete tasks using natural language instructions. This innovation marks a significant step forward in the development of Intelligent Agents capable of seamlessly interacting with humans.
The core of this innovation lies in the use of video tours to train the robots. Researchers film a walkthrough of a specific area, such as a home or office, allowing the robot to “watch” and learn about its environment. With the help of Gemini 1.5 Pro’s extensive context window, the robots can interpret complex commands. For instance, when shown a phone and asked, “where can I charge this?” the robot successfully guided users to the nearest power outlet with a 90% success rate in a 9,000-square-foot area. This demonstrates the remarkable potential of Digital Employees in enhancing everyday life.
Moreover, the study revealed that Gemini-powered robots are not only adept at navigation but can also plan and execute more intricate tasks. In one example, a robot successfully identified and reported the presence of Coke cans when asked if the user’s favorite drink was available. This capability highlights the potential of Non-Human Workers to perform tasks that require a higher level of understanding and planning.
While there are still some challenges to overcome, such as the 10-30 second processing time for instructions, the progress made by Google DeepMind is impressive. As these Intelligent Agents continue to evolve, the prospect of having robots assist with everyday tasks, from finding lost keys to more complex household chores, is becoming increasingly feasible.
Key Highlights:
- Date and Event: On July 11, 2024, Google DeepMind announced advancements in robotic intelligence using Gemini AI.
- Technology: The Gemini 1.5 Pro model enhances robots’ navigation and task completion abilities through natural language instructions.
- Training Method: Robots are trained using video tours of environments, enabling them to understand and navigate spaces.
- Success Rate: Gemini-powered robots had a 90% success rate in completing over 50 user instructions in a 9,000-square-foot area.
- Advanced Planning: Robots can plan and execute complex tasks, such as identifying and reporting the presence of items like Coke cans.
- Processing Time: Current technology requires 10-30 seconds to process instructions, indicating room for improvement.
- Future Prospects: The evolution of Intelligent Agents and Digital Employees promises enhanced assistance with everyday tasks, from simple navigation to complex household chores.
Reference:
https://www.theverge.com/2024/7/11/24196402/google-deepmind-gemini-1-5-pro-robot-navigation
