Nowadays artificial intelligence and robotics operate as largely separate fields.
While AI has impressed us with its decision-making and learning capabilities, it remains within computers.
Robots, on the other hand, have a tangible presence and can execute pre-programmed tasks, yet lack intelligence.
However, this division is gradually dissolving. AI is transitioning from the digital world into the dynamic, physical 3D space, leading to a new paradigm known as physical intelligence.
Understanding Physical Intelligence
Physical intelligence integrates AI’s ability to comprehend text, images, and other online data with real-world machines, increasing their performance. This evolution means AI will no longer stay in our computers but will interact with us in the physical world in different ways. Imagine a supermarket where robots assist you with various tasks, such as carrying heavy boxes. Achieving this requires rethinking machine cognition, design, and learning processes.
The Role of Liquid Networks in AI
For physical intelligence to be effective, AI must operate on computers small enough to fit on a robot. Present-day AI systems can make errors; hence, there is a need for more reliable, compact systems. Daniela Rus, a professor at MIT’s Computer Science and AI Lab, along with her team, has pioneered a new AI approach called “liquid networks.”
Consider a self-driving car, which uses tens of thousands of artificial neurons to navigate. It’s challenging to link these neurons’ activities to the car’s behavior. On the other hand, a liquid network solution uses only 19 neurons, focusing on the road and surroundings. This reduced complexity allows the researchers to understand decision-making processes better. Traditional AI systems use artificial neurons as basic on/off units, whereas liquid neurons employ differential equations to model neural computations more accurately.
The Advantage of Liquid Networks
Liquid networks are not static; they continue to adapt after deployment based on new inputs. This adaptability, rooted in physics models, offers a new way for machines to think. These networks can run on various devices, from smartphones to robots and enterprise computers, opening new possibilities for machine design and functionality.
Innovating with Text-to-Robot and Image-to-Robot
The second critical aspect of physical intelligence involves transforming text and images into robotic actions. Current AI solutions, which rely on statistical models, fall short in this regard. Liquid networks, however, make it possible to convert text and images into functional robots, drastically reducing the time and resources needed for prototyping and innovation.
Human-to-Robot Learning
The third dimension of physical intelligence is enabling machines to learn from human behavior. Daniela Rus and her team have developed a system where humans are instrumented with sensors in a kitchen environment. By collecting comprehensive data on human tasks—such as muscle movement, posture, and gaze—they train AI to teach robots these tasks. This approach leads to robots that can move with agility and adapt to various tasks, from food preparation to cleaning.
Conclusion
The fusion of AI and robotics through liquid networks is leads to huge advancements in physical intelligence.
This technology promises to bring our most imaginative ideas to life, creating intelligent machines that can interact seamlessly with the physical world.
At NextBrain AI, we follow the latest AI innovations to help you extract invaluable business and strategic insights from your data. Schedule a demo today to see the power firsthand.
+34 91 991 95 65
+61 410 497229