Master Thesis: Monocular XR Object Localization with World Models
Core
Investigate if a pretrained world model (e.g., NVIDIA Cosmos) can use semantic and scale priors to improve metric localization of distant moving objects from monocular XR video.
Role type
Master Thesis Researcher (XR Perception)
Builds
Monocular XR object localization system using world models
Domain
Extended Reality (XR), Computer Vision, AI
Deliverable
research
Required skills
World models, monocular depth estimation, semantic segmentation, scale estimation, server-side inference, Quest 3 development
Preferred skills
NVIDIA Cosmos, metric localization, latency optimization
Technologies
Quest 3, NVIDIA Cosmos, RGB-video capture, server-side inference
Responsibilities
Build a Quest 3 RGB-video capture pipeline with server-side inference; Train a lightweight probe to estimate object distance and motion; Evaluate the system against depth-estimation baselines; Assess accuracy, computational requirements, latency, and uplink requirements.
