From road recordings to 3D vector ground truth and simulation-ready assets

Modern AD development needs more than raw sensor recordings. To train, validate, and debug perception and planning logic, teams require a high-precision, machine-readable reference of the road environment. HD maps provide exactly this: a consistent, structured foundation for modern AD workflows.

An HD (High Definition) map is a road network model that captures the structure and topology of roads, lanes, markings, and other traffic-related elements at high precision. In practice, this means lane-level geometry and road semantics represented as clean vector features with a defined coordinate system and quality rules. HD maps can complement sensors by providing a redundant view of the environment, and they can also serve as ground truth for building training datasets through manual and automated labeling of geo-referenced recordings.

aiMotive’s HD map approach

Our key differentiator is that we build and validate the map in full 3D. While many mapping workflows remain effectively 2D – relying on imagery-driven annotations – aiMotive builds and validates every map in full 3D. This model serves as the core building block of a road-level digital twin: a structured, measurable representation of a real location that can be used across simulation, validation, and data generation workflows.

HD maps’ value show up in engineering workflows where you need repeatable, measurable reference data:

• Training data generation: Automatically annotating large volumes of recordings by overlaying HD maps onto the recorded sensor data using high-precision localization. Using this technique, you can harvest training data from the selected sections indefinitely, under various weather, lighting, and traffic conditions. 

• Simulation and scenario creation: Recreating real-world locations with high-fidelity road geometry and semantics to support closed-loop testing. 

• Validation and metrics: consistent map features enable measurable validation, regression testing, and KPI computation across releases.

• Digital twin development: combining precise 3D road geometry and semantics to create a reusable, simulation-ready representation of a real environment.
   

Many mapping outputs look fine until you try to use them in a real workflow. Urban intersections are a typical stress test.

That is why topology checks and junction logic matter. Our process emphasizes:

• Continuity through intersections

• Correct connectivity and directionality

• Clean geometry that avoids unrealistic bends or artifacts

• Consistent rules that stay stable across the whole mapped area

From data collection to vector ground truth

A reliable HD map starts with reliable inputs. Professional Services can work from customer recordings, but we can also collect data using our own setup. aiMotive offers a lightweight data collection rig equipped with LiDAR and GNSS-INS to gather the raw data needed for mapping.

To ensure local consistency even in GPS-denied environments – such as parking garages – we utilize SLAM-enhanced (Simultaneous Localization and Mapping) workflows. This approach maintains centimeter-level accuracy where GNSS is unavailable, resulting in a robust and reliable data foundation.

HD mapping is not only about “drawing features.” It is also about verification.

We apply multiple QA gates, combining automated checks with human review. Typical validation focuses on:

• Geometry consistency (no drift, gaps, overshoots, or self-intersections)

• Topology consistency (connectivity, directionality, intersection rules)

• Completeness (coverage and expected feature presence)

• Schema correctness (coordinate system, metadata, layer structure)

What we deliver: vectorized road semantics that match real-world road logic

Raw point clouds are merely the starting point. What engineering teams truly need is a structured 3D road model that downstream tools can query, filter, and validate.  

Our GeoPackage deliverables can include a wide set of layers, depending on the customer’s needs. 

Common examples include:

• Lane markings: longitudinal markings such as solid, dashed, double lines and many more that define lane boundaries

• Hatched areas: road surface regions marked with diagonal or parallel paint patterns

• Miscellaneous markings: line-type markings not classified as lane markings (for example stop lines, bike lane separators)

• Road surface signs: point-type pavement markings like arrows, symbols, or text painted on the roadway

• Pedestrian crossings: crosswalk outlines (for example zebra crossings)

• Pavement border: the edge of the drivable road or paved surface

• Pavement type: surface material and structure of roads and paved areas

• Guard rails: steel roadside safety barriers

• Barriers: roadside separation and protection structures

• Lane centerlines: lane center alignment including logical continuation through intersections

• And additional layers on request, including customer-specific schemas and attributes

The goal is always the same: deliver structured map data that can be used directly in downstream workflows, whether that is dataset generation, simulation, or validation.

Integration with the broader aiMotive toolchain

In many projects, HD map creation is one step in an end-to-end pipeline. We can support the full workflow from planning and recording through point cloud processing, vector HD map creation, simulation map build, neural reconstruction, and metrics-based evaluation. In this context, the output can be viewed as a digital-twin pipeline: from real-world capture to a simulation-ready, measurable environment.

HD map creation is one part of a wider workflow. Within Professional Services, mapping can be combined with other capabilities depending on the project scope:

• Virtual asset and map creation in aiSim for recreating locations or extending scenarios 

• KPI computation with aiMetrics on top of collected data 

• Neural reconstruction and rendering for generating simulation-ready 3D environments from recordings

This integrated approach is critical when a project demands more than a standalone file. By aligning HD maps with aiSim’s virtual assets and aiMetrics’ KPI tools, we deliver a complete simulation ecosystem that perfectly mirrors real-world data.

HD maps are only as valuable as their accuracy and usability in real-world engineering. Whether you are focused on dataset generation or high-fidelity simulatuion, aiMotive Professional Services provides the end-to-end expertise – from raw data collection to validated, vectorized ground truth.