Lyra 2.0 builds persistent explorable 3D worlds

The important shift here is that this is not just prettier video generation; it is an attempt to make generative worlds useful as assets for robotics and simulation.

• –Long camera trajectories are the core technical challenge, and Lyra 2.0 explicitly targets the failure modes that usually break them: forgetting previously seen content and drifting geometries over time.
• –The interactive explorer plus 3D reconstruction loop makes the output more than a clip; it becomes a navigable scene that can be revisited from new viewpoints.
• –Export into 3D Gaussians, meshes, and Isaac Sim is the part that makes this interesting for downstream physical AI work.
• –This sits squarely in research, but the product direction is clear: world generation that can feed real simulation and embodied AI pipelines.