NVIDIA showcases a community of robots: they learn and exchange skills like human workers

NVIDIA has demonstrated the advancements of ENPIRE, an experimental platform that combines agent-based artificial intelligence and physical robotics, with the goal of creating machines capable of acquiring new skills directly in the real world. The most surprising demonstration involves a robot that can install a graphics card onto a motherboard, a seemingly simple task for seasoned professionals, but quite delicate and requiring precision and coordination.

The project was shown in a video published on LinkedIn by Jim Fan, NVIDIA's AI Research Director. The footage displays several robots engaged in precision tasks, including organizing small metal pins into dedicated containers and applying electrical ties followed by cutting excess material.

At the core of ENPIRE is an approach called AutoResearch, through which a set of AI agents is given a goal and has access to an infrastructure made up of robots, GPUs, and computing tokens to independently identify the most effective strategy. The agents observe the surrounding environment, make attempts, correct errors, consult technical documentation, analyze previous results, and gradually refine their operational methods.

According to Jim Fan, the project represents one of the first concrete examples of self-training artificial intelligence applied directly to the physical world. The ultimate goal is not only to execute predefined instructions but to create robots that can learn new skills on-site, thereby progressively reducing the need for manual programming by humans.

Tests conducted by NVIDIA also highlight the importance of collaboration among multiple robotic units. A fleet of eight robots can solve complex problems in significantly shorter times compared to smaller groups, as each unit can experiment with different approaches and quickly share the collected information with other machines.

Thus, NVIDIA's ENPIRE goes beyond traditional robotics, which sees a machine perform a programmed task. Essentially, Jensen Huang's company is structuring true communities of robots, capable of interfacing with one another and sharing learned skills just as would occur among normal human operators.