A new study has reportedly showcased Google Gemini's navigating and reasoning capabilities when powering a robot in Google's Deepmind offices.
In the reported study "Mobility VLA: Multimodal Instruction Navigation with Long-Context VLMs and Topological Graphs," the team shows how they used Google Gemini 1.5 Pro to teach a robot to respond to orders and travel about an office.
(Photo : Photo by Tomohiro Ohsumi/Getty Images)
According to Google, the robot had a success rate of almost 90% over more than 50 engagements with employees. Naturally, DeepMind deployed some of the Every Day Robots sitting around since Google shut down the project amid significant layoffs last year.
In a series of project-related films, DeepMind staffers begin with a smart assistant-style "OK, Robot," before instructing the system to complete various chores within the 9,000-square-foot office area.
In one case, a Googler asks the robot to take him somewhere to sketch things. "OK," the robot says, sporting a jaunty yellow bowtie. "Give me a minute. The robot guides the user to a wall-sized whiteboard, saying, "Thinking with Gemini." In a subsequent video, another human instructs the robot to adhere to the instructions on the whiteboard.
A basic guide explains how to get to the "Blue Area." The robot pauses momentarily before taking a lengthy path, eventually leading to a robotics testing location.
Read Also: Tern AI Develops Low-Cost GPS Alternative Amid Growing Security Concerns
Navigation Training Through AI
Before these films, the robots were acclimated with the workplace using what the team refers to as "Multimodal Instruction Navigation with Demonstration Tours (MINT)." This entails taking the robot about the office while pointing out various landmarks with voice.
Next, the team employs hierarchical Vision-Language-Action (VLA) to link environmental knowledge with common sense reasoning ability. Once the procedures are completed, the robot will be able to respond to written and drawn orders and gestures.
AI-Powered Navigation for Jets
Artificial intelligence, in general, continues to be leveraged as an utterly effective navigator in various industries. US AI fighters are now also being developed to navigate the skies without GPS.
The US's current objective is to create additional AI technology to handle its fighter planes, including three crucial aspects that would give it a competitive advantage.
According to reports, the country's military is now focused on major advances that will enhance technology and offer capabilities like flying without GPS and learning new tactics.
It was discovered that relying on GPS has drawbacks, such as utilizing disruption technology to blind the pilot and inhibit communication. However, being GPS-less on aircraft and relying on AI to sense the Earth's magnetic fields might prevent such interruptions.
On one hand, AI's algorithm only understands what has been uploaded or applied to it on the ground, and flying requires it to rely on stored information. Having said that, testing may utilize its data to inform algorithms while preparing for future endeavors or activities.
Related Article: AI Drones Leveraged to Detect Underwater Mines by US Navy
