Autonomous Driving using High-Definition Maps and a Cloud Data Platform

Utilizing a streaming file system, all road picture data is pooled, processed, and filtered for additional validation and intelligent analysis using machine learning models.

FREMONT, CA: Automatization has four degrees in vehicle engineering: first, manufacturing is controlled by machines configured by people; second, real-time production is the key feature; third, decentralization; and fourth, industrialization with no human participation. There are financial advantages, such as lower production costs and a stronger presence in a competitive market.

The production of 3D HD Maps, from eliciting source data to generating and releasing the maps themselves, is crucial for Fully Automated Driving (FAD). This is an activity where artificial intelligence can help. The process of creating a map is organized into five basic stages:

Real-Time Streaming Perception

Using vehicle-mounted sensor devices, an Edge Perception stack is constructed to enable HD map observations and crowd-sourced updates. The method entails real-time detection of road elements in video feeds, and this dynamic information is then used to create self-healing maps.

Data Aggregation, Processing, and Filtering

Massive datasets are handled by engineers from various sources, including core maps and on-vehicle sensor systems. Utilizing a streaming file system, all road picture data is pooled, processed, and filtered for additional validation and intelligent analysis using machine learning models. All data can be ingested in an NDS (Navigation Data Standard) format to eliminate vendor lock-in and maintain compatibility between systems.

3D Map Creation and Maintenance

Live 3D maps are created and updated in real-time, allowing for exact positioning for lateral and longitudinal vehicle control. These multi-layer maps contain information at multiple levels (roads, lanes, lane groups and individual lanes, geometry) and are regularly updated with new road properties like signs, markers, crosswalks, bicycle lanes, and objects.

Map Delivery                                                      

Customers can download data-intensive HD maps as a geographically tiled and functionally layered data solution appropriate for cloud consumption by direct-to-vehicle and OEM (Original Equipment Manufacturer). These maps are also a vital source of information for the client's numerous in-vehicle software development programs.