Protege, the trusted source for high-quality, real-world training data curated for AI development, today announced a partnership with Grab, Southeast Asia’s everyday superapp, to help make selected real-world data assets from GrabMaps available to AI developers through the Protege platform.
Through this partnership, AI teams will be able to access anonymised data that helps models understand real environments as they are experienced in the world, not isolated objects alone or simulated scenes. The collaboration adds grounded regional context to Protege’s catalog, while supporting the development of world models, geospatial systems, robotics, and other AI applications that need to reason about real places and changing conditions.
Data for grounded spatial intelligence
As models continue to evolve, they need data that reflects how places are laid out, how conditions change, and the real-world variation – and randomness – that models encounter outside controlled settings.
In particular, GrabMap’s expansive location data assets naturally support model-builders in the following areas:
- World models and spatial reasoning: Real-world environmental context for systems learning how places, objects, and conditions relate to one another.
- More representative regional coverage: Grounded perspectives from Southeast Asia, a region that remains underrepresented in many Western-centric AI datasets.
- Temporal and operational understanding: Context that can help models reason about change, movement, and real-world conditions over time.
- Grounded evaluation: Location-linked information that can support spatial benchmarks and evaluation against real-world ground truth.
- Multimodal AI development: Complementary data sources that can be combined across different use cases.
World models and robotics need dense, diverse recordings of how environments and humans interact. Protege’s existing catalog spans human-, robot- and object interactions; Grab’s data adds the environment layer that connects those modalities to the places in which activity unfolds.
In addition, GrabMap’s data is recaptured routinely, creating a regularly refreshed representation of roads and places. This matters especially in Southeast Asia, where mixed-traffic streets, informal markets, tropical weather, and non-Latin scripts remain chronically underrepresented in Western-centric training data. The imagery, road topology, points of interest, and traffic context are linked by geography so multimodal by construction. This makes it authentically fusible rather than something researchers have to stitch together from separate sources.
Together, these characteristics give AI systems a stronger foundation for understanding not just what the world looks like, but how it is organized, how it changes, and what actions are possible within it. Data collected from real environments can help close the gap between models trained in controlled settings and the complexity of the physical world.
“AI systems need to understand the world as it is, not just as it appears in isolated or simulated data,” said Hesson Mak, Grab’s Regional Manager, Business Development for GrabMaps. “Grab’s mapping platform operates across some of the world’s most diverse and dynamic environments, giving GrabMaps hyperlocal coverage across Southeast Asia and a continuously refreshed view of how places and conditions change. Powered by Grab’s real-world ecosystem, these data assets provide a valuable foundation for models learning about real places and real conditions. We’re pleased to partner with Protege to help make this context available to AI developers for a thorough, responsible path for training and evaluation.”
“GrabMap’s purpose-built mapping operations and participating driver-partners have helped produce a remarkable record of the physical world,” said Grant Murphy-Herndon, Protege GM of Spatial & Physical Intelligence. “Grab gives us real-world footage of streets, paired with location and road context, refreshed in a timely manner across hundreds of cities. This supports world models learning how real environments actually behave, vision-language model understanding of Southeast Asia, and researchers getting a spatial benchmark they can verify against real-world data. We think this is a meaningful step forward for spatial and physical intelligence.”
Through the partnership, Protege will support a clear, licensed path for bringing selected GrabMaps data assets to AI model builders. The collaboration is designed to help AI teams access fit-for-purpose data for training and evaluation while preserving the governance, licensing clarity, and responsible data practices expected by data holders and model builders.
About GrabMaps
GrabMaps is Grab’s Southeast Asia-focused mapping and location data provider, helping businesses build more reliable location-enabled products and improve everyday operations across the region. Its capabilities include places and address data, road and traffic information, street-level imagery and proprietary mapmaking tools. Built from Grab’s experience operating across eight Southeast Asian markets, GrabMaps combines local knowledge, continuously maintained data and real-world operational insights to support use cases such as routing, logistics, search, navigation and infrastructure planning.
About Protege
Protege unlocks the right data, purpose-built for AI. The company provides access to the richest real-world data across domains spanning healthcare, media, and more. Beyond data licensing alone, Protege drastically shortens the time from raw data to purpose-built datasets for training and evaluation.
The company works directly with leading AI labs, robotics companies, and frontier model developers to source, refine, curate, and operationalize high-value datasets for training and evaluation. Across video, media, and other domains, Protege helps data and content rights holders turn high-quality, underused data into new AI revenue streams without giving up control.
To learn more about how your organization can unlock new revenue by ethically licensing your content for AI, or access datasets purpose-built for AI, visit https://withprotege.ai/data-providers.