NVIDIA DRIVE Map
NVIDIA DRIVE Map™ is a multi-modal mapping platform that combines ground-truth and fleet-sourced map engines to achieve accuracy and scale. It serves as a foundational dataset for labeling, training, validation, and 3D environment reconstruction for simulation. DRIVE Map also includes four localization layers, giving developers the flexibility to integrate into any autonomous vehicle system.
Two Map Engines
DRIVE Map is built on two map engines: ground-truth and fleet-sourced. The ground truth map engine provides the accuracy required for high levels of autonomy and improves the data pipeline. The scalability of fleet-sourced maps, powered by MapStream, removes geographic restrictions for AV feature development. With the map confidence indicator, the vehicle can determine the quality of the map data to use for localization and planning.
Open MapStream
MapStream is the perception data protocol to communicate road observations between the vehicle and the cloud. It contains landmarks from the perception DNNs such as traffic lights, signs, poles, wait lines, lane dividers, driving paths, road boundaries etc., as well as geospatial coordinates (WGS84) and vehicle trajectory from GNSS/IMU/CAN. MapStream API is open — AV developers can leverage NVIDIA DRIVE Perception, or encode their own perception data in the MapStream format to contribute to map creation and update.
Multi-Modal Localization
DRIVE Map has four redundant localization layers — camera, radar, lidar, GNSS — built using rich sensor data from DRIVE Hyperion data collection vehicles and partner consumer fleets. Multi modality in localization maps provides developers with the flexibility to use sensor data independently or fuse various layers together based on the redundancy level planned for their AV systems.
Ground Truth Data Set
The ground truth map is robustly validated, providing a highly accurate and reliable dataset. It enables auto-labeling by projecting perception data into the map. This ground truth data serves as the foundation for DNN training and fleet-sourced map validation. It can also be used to create drivable environments for simulation. The versatility of this map improves the data pipeline and accelerate development and time to market.
Resources
Peek under the hood to experience NVIDIA’s latest autonomous driving innovations through DRIVE Labs and DRIVE Dispatch.