Google DeepMind releases Perceptor-XL: a foundation model for spatial understanding and AR design

AI · 6 min read

Perceptor-XL is DeepMind's latest foundation model focused on depth estimation, surface reflectance, and object permanence for AR applications. Trained on dense multi-view datasets and synthetic scenes, the model outputs layered scene graphs, depth maps, and lighting probes suitable for placing virtual elements realistically.

The model integrates with AR SDKs to provide placement suggestions, occlusion masks, and suggested materials that match the scene's lighting. Designers can use Perceptor-XL inside prototyping tools to preview how virtual typography or UI would read against complex backgrounds.

DeepMind emphasized that the model is optimized for streaming inference on mobile GPUs and includes a small on-device runtime for low-latency interactions. The release includes sample projects for Unity and Android to accelerate integration into AR design workflows.