DeepScenario's summer brief
Dear friends and partners of DeepScenario,
As the summer season gets into full swing, we want to share a look back on some of our highlights from an incredibly productive first half of 2026 with you.
Over the past couple of months, we have significantly advanced our scenario reconstruction capability for dashcam footage and are using this technology, for example, to virtualize large-scale fleet data for a leading OEM. We have also released our first dataset from this perspective on our web app.
Additionally, we have successfully introduced the generation of high-quality end-to-end training samples designed to address AV2.0 use cases. Our approach creates 3D bounding boxes and occupancy grids as well to make the end-to-end system more interpretable. We also proved the robustness of our 3D perception algorithms across different challenging weather conditions, including night and snow. Lastly, we have unlocked even higher levels of realism in simulation: Users can now leverage contour-accurate 3D vehicle meshes generated directly from video and run our real-world trajectories within high-fidelity static environments.
Beyond our applications in the automotive market, this half of the year also marked an exciting milestone as we started bringing our advanced spatial AI capabilities into the defense sector. One of our core use cases is autonomous vehicle navigation in highly complex, unstructured terrains, comprising real-time 3D occupancy perception, object tracking, and scenario reconstruction.
We also enable surveillance applications through 3D tactical object tracking using ordinary RGB cameras, as well as centimeter-accurate 3D environment reconstruction combined with 3D object detection and tracking from standard drone data for aerial surveying. To explore these specialized solutions in more detail, feel free to take a look at the newly launched defense page on our website.
Moreover, our commitment to advancing the state of the art remains stronger than ever, highlighted, for instance, by our recent contribution to a paper that was honored with the best paper award at CVPR’s DriveX Workshop. Researchers from CARIAD and KIT also leveraged our precise drone data in their publication on training highly realistic, computationally efficient multi-agent behavior models. Finally, we got to experience an amazing moment at the STADT:up final event: CARIAD delivered an impressive live driving demo on the test track, running a prediction model that was trained exclusively using simulation and drone data provided by DeepScenario.
After this amazing first half of the year, we are already looking forward to the next months and the exciting developments we will be able to share with you. To stay in the loop about our mission and progress, come follow us on LinkedIn and never miss an update!
Wishing you a fantastic summer
Your DeepScenario Team