LIMITLESS VIRTUAL WORLDS / PHYSICAL AI
Rebuild reality
from a single drive.
Netropy reconstructs recorded driving into living 3D worlds you can re-run, edit, and stress-test — closing the loop on autonomous driving validation without another mile of real-world risk.
Backed & built with
The premise
Your fleet has already seen the edge cases. They are sitting, frozen, in recorded video. We bring that data back to life as a world you can drive again — and again, differently.
From 2D to 3D
Two routes from a real drive to a virtual one.
Start with footage you already have. Pull the scenario out of it, then rebuild the world or regenerate the drive — whichever reaches validation faster.
Input
Any drive
dashcam · multi-cam · fleet log
SIGNAL INExtract · Scenario
GenixScenario
Formerly GenixCore
Mine edge-case scenarios out of any driving video.
GenixScenario reads ordinary footage in three dimensions — reconstructing the environment, recovering ego-motion, and tracking every actor — then hands you a clean, taggable scenario ready for your validation tool-chain.
- 3D environment reconstruction from camera-only input
- Dynamic object detection, classification & tracking
- Searchable scenario metadata & tagging
- Export to ASAM OpenSCENARIO & OpenDRIVE for any simulator
Reconstruct
GenixRecon
Turn a logged drive into a living 3D world.
GenixRecon rebuilds recorded driving logs as a dynamic 3D Gaussian-splat world — geometry, appearance, and every moving actor, faithfully recovered over time. The result drops straight into GenixSim, ready to re-render from any viewpoint.
- Dynamic 3D Gaussian splatting from raw driving logs
- Full scene and per-actor motion recovered over time
- Free-viewpoint, photoreal neural rendering
- Sim-ready output that hands off directly to GenixSim
Simulate
GenixSim
Edit the world. Close the loop.
GenixSim turns a reconstructed world into an interactive testbed. Reshape the scene by hand, by config, or by asking an agent — then drive your full stack inside it, closed-loop, on high-fidelity generated sensor data. A simulator built for autonomous driving, not adapted to it.
- Edit scenarios three ways — manual, config, or LLM agent
- Closed-loop test, training & validation for E2E / VLA stacks
- High-fidelity generated camera & sensor streams
- Focused on the road, where general robotics sims stop short
Generate
GenixRetake
Re-drive the scene. Skip the rebuild.
GenixRetake edits the drive itself. Move the ego vehicle or any surrounding actor onto a new path, and a video model renders the alternate take directly from the original dashcam — no full reconstruction, no hand-labeled scene. The fastest route from “what if” to footage.
- Re-route ego and surrounding objects on a dashcam clip
- Direct video generation — no reconstruction step
- Works without full scene annotation
- New corner cases delivered as ready-to-use video
Where it runs
ADAS / L2 validation
Convert fleet driving into scenario libraries your existing simulators can consume today.
E2E & L2+ autonomy
Reconstruct and re-simulate to validate end-to-end and VLA driving models against reality.
Robotics
Generate virtual environments and synthetic data to develop and tune robot policies — including on NVIDIA Isaac Sim.
The team
Engineers who have put autonomy on the road.
Netropy is a team of engineers and researchers with years of experience in AI and autonomous driving — building perception and simulation that reached production with global OEMs. We have lived the validation problem from the inside, and we built Netropy to solve it properly.
A research-driven team out of Seoul National University, KAIST, and EPFL — backed by SpringCamp, selected for TIPS, and an NVIDIA Inception partner.
We're hiringStart the loop
Bring us a drive.
Leave with a world.
Tell us what you need to validate. We will show you the same scenario, reconstructed and re-simulated from your own data.
Talk to us