Waabi has introduced STRIDE, an AI method for reconstructing changing driving scenes from camera and LiDAR data. Its practical promise is more than a lifelike replay: the system can separate moving objects so researchers can edit a scene, changing an actor’s path or adding and removing actors for simulation.
Waabi Watch analysis
What happened
Waabi describes STRIDE as a self-supervised method that builds a 4D reconstruction in a single forward pass. It combines image features with LiDAR measurements, then uses a 3D Point Transformer to estimate scene geometry and motion. Learnable instance tokens group moving parts into distinct actors without manual annotations.
Waabi says tests on the Waymo Open Dataset showed state-of-the-art performance for scene reconstruction and motion estimation. Its examples include removing an actor, changing an actor’s trajectory and inserting an actor from another scene. These are research results and demonstrations reported by Waabi, not evidence that the system is deployed in a vehicle.
Why it matters
Self-driving systems need more than realistic-looking scenery. Useful simulation also needs coherent motion and the ability to ask “what if that pedestrian had crossed here?” STRIDE’s combination of reconstruction and actor-level editing is aimed at making those scenarios easier to create and vary.
That could make simulation data more flexible for training and testing autonomous systems. The paper does not establish how much faster or cheaper this makes that work in practice, but it tackles a real bottleneck: manually annotating and editing dynamic scenes is a tedious way to build a large library of edge cases.
Our read
The interesting step is tying scene reconstruction to editable, object-level structure, rather than producing a pretty digital twin and leaving it stuck in place. If the method generalises beyond the reported benchmarks, researchers could change a scene’s actors and motion without rebuilding the whole thing by hand. That is a useful research direction, with the usual gap between a compelling demonstration and a tool ready for everyday use.
What to watch
- Whether the STRIDE paper and code make the method reproducible.
- How it performs on datasets beyond the Waymo Open Dataset.
- Whether its scene edits translate into measurable gains for autonomous-driving simulation.
Discussion spark: For autonomous-driving simulation, which matters more: photorealistic scenes, or the ability to reliably edit actors and their behaviour?
Sources and evidence
- STRIDE: Towards Self-Supervised, Generalizable, and Decomposable 4D Driving Scene Reconstruction (10 November 2026, 00:00 UTC)
Independent WittyWires tracker for public updates about Waabi. Not affiliated with or endorsed by Waabi; this is not an official account.