Open-world

A setting where the environment is uncontrolled and things absent from training keep appearing, and the system must still work.

“Open-world” contrasts with the “closed-world” assumption, under which the categories and scenes encountered at test time are assumed to all fall within the training range. In 2014, Abhijit Bendale and Terrance Boult formally defined open-world recognition in visual recognition, requiring a system to detect unknown categories, label them as unknown, and gradually learn them over time. In embodied AI, the term is used more broadly, referring generally to leaving the lab and entering real, uncontrolled settings such as homes, shops, and the outdoors, where objects, layout, lighting, and human behavior cannot all be enumerated in advance. Physical Intelligence's π0.5, released in 2025, specifically targets open-world generalization, demonstrating long-horizon tasks such as tidying a kitchen or bedroom in entirely new homes. This usually depends on diverse training data and knowledge transferred from the internet.

Exampleπ0.5 is placed in a home that never appeared in its training data and, given an instruction such as “tidy up the kitchen,” completes the multi-step cleanup on its own.

Also called

In-the-wild

Related

Open-vocabulary · Out-of-Distribution · Scene Generalization · Long-tail Problem · π0.5 · In-the-wild Data

Sources

Towards Open World Recognition
π0.5: a Vision-Language-Action Model with Open-World Generalization

As of

2025-04