We’re building a scalable Spatial AI pipeline that supports foundation models for semantic segmentation, object detection, scene understanding, and intent prediction.

The full-stack pipeline covers on-device data collection, cloud-based annotation tools, APIs/SDKs, and partnerships spanning academia, startups, and enterprises.

The goal is to enable 3D world understanding within AI systems through a cohesive pipeline and broad cross-sector collaboration.

The project aims to become the world’s largest open-access driving dataset, targeting about one million 30-second clips, with plans for future additions like 2D/3D boxes, scene graphs, behavior cues, and diversity metadata.

Backing this effort is a decentralized network of contributors, more than 20 million kilometers of road coverage, and over 3,500 deployed devices.

The dataset is released under a permissive non-commercial license, with plans for extended versions for commercial partners to support open infrastructure and collaborative AI development.

Licensing emphasizes non-commercial use upfront while keeping pathways open for commercial collaborations to democratize access to real-world data.

Access to the dataset is available at rovr.network/#/dataset for researchers, developers, educators, and innovators.

The dataset is now accessible to the wider community, inviting researchers, developers, educators, and innovators to participate in shaping Spatial AI.

ROVR’s DePIN model emphasizes scalability and global reach, with incentivized participation through token rewards as the network has already achieved substantial road coverage and device deployment.

The initial release comprises 1,500 fully synchronized clips totaling over 1 terabyte, spanning urban, suburban, highway, and special scenarios like construction zones and school crossings.

Data in the initial release includes LiDAR point clouds, front-facing RGB video, high-frequency IMU data, centimeter-level RTK GPS, and anonymized scenes.