New MBZUAI AI framework to fast-track robot learning
New framework could dramatically improve robotic performance over alternatives
#UAE #education - Abu Dhabi research-focused university Mohamed bin Zayed University for Artificial Intelligence (MBZUAI) has co-authored breakthrough research published in Nature Machine Intelligence introducing the Tactile Skills AI framework, enabling robots to rapidly master complex physical tasks with near-100% success rates across 28 industrial applications. UAE Vice President for Research Professor Sami Haddadin collaborated with researchers from University of Sussex and Imperial College London to develop the embodied AI system that bridges human expertise with robotic capability. The new AI framework could potentially transform automation across manufacturing, logistics, and healthcare sectors by reducing setup times and costs whilst dramatically improving performance over traditional deep-learning approaches.
SO WHAT? - The new Tactile Skills AI framework could prove to be a fundamental shift from trial-and-error robot learning to structured skill acquisition inspired by human vocational training. According to the researchers, the framework's ability to achieve industrial-grade performance without massive datasets addresses critical bottlenecks that have limited robotic deployment in delicate manufacturing processes. Therefore, this could play a key role in accelerating automation adoption across sectors where precision handling of flexible materials and complex assembly operations remain challenging.
Here are some key points about the new framework:
Professor Sami Haddadin at Mohamed bin Zayed University for Artificial Intelligence (MBZUAI) has co-authored research on a new Tactile Skills AI framework for robotic learning, together with former PhD student Lars Johannsmeier and colleagues from University of Sussex and Imperial College London. A paper for the new research was published in Nature Machine Intelligence on 23 June 2025.
The Tactile Skills AI framework achieved near-100% success rates in tests across 28 distinct industrial tasks including plug insertion, precision cutting, and complex assembly operations, even when encountering unexpected changes in positioning or environmental conditions.
As a result of training via the new framework, robots successfully assembled complex industrial devices used in bottle-filling plants, demonstrating practical applicability for real-world manufacturing scenarios requiring delicate tactile manipulation and precision assembly capabilities.
The system combines expert process knowledge with reusable tactile control components, dramatically reducing energy consumption and improving performance compared to traditional deep-learning approaches that rely on extensive trial-and-error learning.
The AI framework enables operators without deep robotics expertise to deploy robots across diverse tasks, significantly reducing setup times and costs whilst maintaining industrial-grade performance standards and reliability.
The new taxonomy-based approach creates structured curricula for robots similar to human vocational training programmes, facilitating rapid skill acquisition and transfer across different applications and manufacturing processes.
Research addresses longstanding challenges in robotic manipulation where traditional automation has struggled with reliably performing delicate, tactile tasks involving flexible materials and precision assembly operations.
MBZUAI launched its master’s and Ph.D. programmes in robotics in 2023.
[Written and edited with the assistance of AI]