Building the world inside aiSim

Written by Bence Czanik, Laszlo Kun / Posted at 10 June 2026

Building the world inside aiSim

How we create the 3D environments that make simulation credible

A simulator is only as useful as the world it runs in. Sensor models, physics engines, and scenario logic all depend on one thing being right first: the environment. At aiMotive, we've built a full in-house 3D content pipeline that feeds aiSim with everything it needs – from individual curb stones to multi-kilometer highway networks across three continents.

Here's how it works.

The Asset Library – the raw material for everything

Before a single map or vehicle enters aiSim, it starts here. Our Asset Library contains over 5,000 individual 3D assets – models, textures, and materials – all built in-house by our graphics team.

It covers everything a road environment needs:

• Road furniture: lamp posts, barriers, signs, traffic lights – and yes, individual tufts of grass

• Reusable surface materials: asphalt, concrete, soil, grass, sand

• Vehicle accessories: police light bars, taxi signs, learner driver plates

• Pedestrian props: bags, umbrellas, phones

Because aiSim users receive the full Asset Library as part of their deployment, IP hygiene is non-negotiable. Every asset is original — no third-party store models, ever.

The toolchain is the same as VFX and game development: Maya, Blender, Houdini, Adobe Creative Suite, increasingly supplemented by generative AI tools like Rodin and Meshy. Everything is managed inside Unreal Editor – not as a renderer (aiSim uses its own engine), but as a content platform built for exactly this kind of scale. A custom plugin handles export into aiSim's native format.

Maps – two layers, one environment

Every map in aiSim is built in two phases.

Phase 1: Procedural generation from HD maps

The starting point is always an HD map – a detailed vector dataset describing lane markings, road edges, crossings, curbs, and traffic infrastructure. Most commonly, our annotation team produces these from LiDAR scans.

Our internal tool, Atlas (included in the aiSim package, so customers can generate their own maps), converts this data into road surfaces, lane markings, pavements, and terrain – automatically. Where the source data is rich enough, guardrails and traffic signals are placed automatically, too.

Phase 2: Manual decoration

The graphics team then takes the procedurally generated base and adds everything else. Working large to small:

• Terrain: Sculpted using real-world elevation data, often sub-metre resolution

• Buildings: Placed individually from the library, or generated procedurally using our façade module system – floor plans can come from the HD map or from Open Street Map

• Props and vegetation: Placed using custom internal tools (fence generators, telegraph pole connectors, parametric traffic light placers) and Unreal's foliage paint system

Any asset needed for a map that isn't yet in the library is modeled and added – so the library grows with every project.

World Extractor – the fast lane

For environments where speed matters more than manual precision, our newest tool World Extractor generates Gaussian Splatting maps directly from video footage and LiDAR scans. What was recorded is what you get – fast. The two approaches can be combined: GS environments can be augmented with Asset Library elements to fill in the gaps.

Current map portfolio: 250+ km of road, 50+ maps – urban, highway, parking – across the US, Japan, and Europe.

Vehicles – from model to simulator

Most vehicles are based on models purchased from specialist automotive modeling partners. We also process customer-supplied CAD data, and have occasionally 3D-scanned physical vehicles directly.

Every vehicle goes through the same pipeline:

• Optimisation – retopology to ~150,000 triangles if the source model is denser

• Node hierarchy– component naming and structure to match aiSim's expectations, so wheels rotate, and the physics system knows what it's working with

• Light setup – emissive surfaces and virtual light sources configured for headlights, indicators, and brake lights

Current library: 90+ cars, including buses, trucks, construction vehicles, and emergency services.

Pedestrians – rigged, animated, and diverse

Pedestrian models are sourced from specialist suppliers and built in-house. This is where generative AI tooling has had the most visible impact on our workflow – accelerating both geometry creation and texture painting significantly, without replacing the digital sculptor.

Once the static model is ready:

• Skeleton & skinning: Weight painting defines how the mesh follows the bones – mostly manual, and it shows in the quality

• Animation: Motion capture data, retargeted to each character – two base animations per pedestrian: idle and walk

Current library: 200+ characters – male, female, children, wheelchair users, scooter riders – across European, Asian, and African profiles. Plus animals.

​​​​​​

The engine underneath

The content layer described above runs on top of aiSim's simulation core – and that matters as much as the assets themselves.

aimSim is the world's first ISO 26262 ASIL-D certified automotive simulator, meaning simulation results can be part of a formal safety case – not just a development aid. Sensor outputs are physically accurate down to the pixel level, from camera response curves to radar cross-section modeling. And with neural simulation, real-world recordings can be used to generate sensor-accurate synthetic data that is indistinguishable from the original footage – which is exactly what powers World Extractor.

This combination – a certified engine, physical accuracy, and neural reconstruction – is what separates simulation-as-validation from simulation-as-testing.

Why the content layer is a competitive advantage

Simulation fidelity is often discussed in terms of sensor physics or scenario logic. The content layer gets less attention – but it's what determines whether the scenarios you're validating against actually resemble the world your software will face.

A simulator with a sparse or geographically narrow environment library will systematically underrepresent the edge cases that matter. The depth and variety of aiSim's content – and the tooling that lets customers extend it with their own HD map data and World Extractor captures – is what makes validation meaningful rather than theatrical.

Interested in what our simulation tool can do for your program? Visit our aiSim subpage.