A paper published by Sami Haddadin, VP for Research at Mohamed bin Zayed University for Artificial Intelligence (MBZUAI), and his research team, has introduced a groundbreaking embodied AI concept known as ‘Tactile Skills.’
These skills are a theoretically sound, practical, and scalable framework, inspired by the human neural system and human vocational training.
In essence, the method represents a specialized curriculum for robots, allowing them to rapidly learn and master new physical tasks.
One important advantage of this method is that it doesn’t rely solely on extensive trial-and-error or massive datasets as opposed to a traditional machine-learning (ML) method. Instead, it blends expert process knowledge with reusable tactile control and adaptation components, significantly simplifying and speeding up the robot’s learning process.
When tested on 28 distinct industrial tasks, including complex operations like plug insertion and precision cutting, the robots were able to achieve a near-100% success rate and industrial-grade performance, even when encountering unexpected changes in object positioning or environmental conditions. “This research represents an important leap toward widespread automation, transforming robots from specialized tools into adaptable, skilled assistants and, ultimately, physical AI agents,” explained Haddadin.
