So let's say that if we think about using this reconstruction, as a way that i would assume is to train ai right and this could be a car walking in a street a robot understanding an indoor environment what are the kind of like use cases are these space used for some of the spaces used for that are they good enough can you tell about that specific view of how you intend yeah let's say this reconstruction to be used

You know, so Gaussian splatting is just a technology, right? And it's a very cool medium for understanding the physical world because it gives you photorealistic representation of the splats.

It gives you an immersive view.

If you place the virtual camera inside the scene and you collect, like you record the video from inside the simulation, it's going to be almost indistinguishable from the real video collected from the real world.

that's mind-blowing i mean that's like almost like a matrix and one of the advantages that you can create as many trajectories as you want within the simulation so you collected data once and then you run like thousands and tens of thousands different scenarios inside the scene and it would feel like a real environment so for the robot it wouldn't make any difference whether this data was collected from the real world versus the simulation, except maybe moving parts, right? So that the parts are not moving, it's just static.

So that's just the video layer.

I mean, this is just the image layer, the perception.

And then there is that big, far away challenge that seems to be where really the big goals are of this technology you as splatica where do you see navigating these two objectives because maybe one just revolves around the visual quality and the speed of which goes and the other is like okay we need to build our own visual model and refine it and annotating as a startup right how do you navigate that