Apple received 3 granted patents in H1 2026 across 2 categories: AI & Machine Learning (1) and VR & AR (2).
The VR & AR patents cover technical foundations for enhanced reality experiences, including an agent-agnostic motion planner that generates movement paths for virtual characters across different XR applications and a frame rate extrapolation system that uses motion sensors to generate intermediate frames for reduced latency. The AI & Machine Learning patent describes an image reprojection technique that applies deep learning to predict and reconstruct frames, reducing rendering costs in AR and VR headsets.
Note: The intro provided contains some categorization errors (the motion planner is listed under AI & Machine Learning in the data, not VR & AR, and the image reprojection patent is listed under VR & AR, not AI & Machine Learning). The paragraphs below follow the patent data exactly as provided.
---
1 AI & Machine Learning patent covers motion planning for virtual characters in XR and VR environments. Rather than requiring developers to build a custom motion planner for every character or agent they create, the system analyzes each motion controller's characteristics at runtime and generates movement paths that are compatible with that controller's specific capabilities and control scheme. It produces verified plans using rapidly-exploring random trees alongside machine-learned root motion models, adapting on the fly to how each individual agent moves without locking developers into agent-specific solutions.
The 2 VR & AR patents both address the challenge of rendering smooth, low-latency visuals in headsets, approaching it from different angles. One patent describes a frame rate extrapolation method that uses motion sensor data to synthesize intermediate frames, and its core contribution is a hybrid depth-processing approach: planar homography handles the central field of view while per-pixel homography (which accounts for variable depth) covers the periphery, with a blending function stitching the two together to reduce flickering without the computational cost of full per-pixel processing across the entire display. The other patent applies causally-constrained deep learning models to temporal supersampling, meaning the system predicts and reconstructs frames without drawing on data from future frames, a constraint that keeps artifacts out of the output while preserving the real-time performance that head-tracked XR displays require.
All data sourced from USPTO patent filings. Google Patents may take several weeks to index recent publications. If a link is unavailable, search for the patent number at USPTO Patent Public Search.