It has become known that artificial intelligence helps robots become more humane

Scientists have presented an innovative platform for human-robot interaction based on the architecture of transformers and Retrieval-Augmented Generation (RAG). This solution gives machines the ability to learn from mistakes and adapt instantly.

Researchers have created a progressive methodology for organizing cooperation between people and robotics in industrial conditions. The core of the development is a combination of Retrieval-Augmented Generation (RAG) and specialized transformer neural networks. This architecture provides robots not only with access to current data in real time, but also with the ability to flexibly correct their actions. The scientific results are published in the journal Scientific Reports.

The proposed methodology allows automated systems to analyze erroneous actions and promptly correct behavior, which is critical for production cycles. The key element is the regret-based learning mechanism, designed to study errors and prevent their reproduction.

Pilot tests in a virtual production environment recorded significant improvements in key metrics. Operation times were reduced by 28.5%, failure rates were reduced by 60.2%, and the need for operator adjustments was reduced by 79.7%. The interaction “intimacy” factor, reflecting the level of trust and frequency of interventions, was assessed separately.

The development plays a decisive role in the transition to the Industry 5.0 concept, where adaptability and flexibility are central requirements. At the same time, the authors note existing technological barriers, including problems of scaling, personalization of learning systems, and the ethical aspects of autonomous decision-making.

Read also: Super-profits of AI developers revealed

MK in MAX: main news - fast, honest, nearby

Authors: