This is like a GPS for inside buildings: you point a camera around a room, and the system figures out exactly where you are on a 2D floor plan by using smart 3D understanding of the space.
Traditional indoor localization (finding a person/device’s position inside a building) is expensive, requires special beacons/sensors, or is too inaccurate. Architects, facility managers, and property technology teams struggle to reliably align what a camera sees with existing floor plans for navigation, inspections, and asset tracking. PALMS+ aims to robustly match camera images to floor plans using AI that understands depth/geometry, without heavy infrastructure.
If matured, the moat would come from robust geometric localization performance across many building types and lighting conditions, plus any proprietary training datasets of real buildings and floor plans. The modular design also allows swapping in better depth foundation models over time, keeping performance competitive.
Hybrid
Unknown
High (Custom Models/Infra)
Running depth foundation models and geometric matching at scale on edge/mobile devices may be constrained by compute and latency; robustness across diverse real-world building layouts is another challenge.
Early Adopters
Unlike generic AR kits and visual-inertial odometry focused on SLAM and tracking (e.g., ARCore/ARKit), PALMS+ is specifically targeted at aligning monocular images with existing 2D floor plans using depth-aware foundation models in a modular fashion. This makes it more relevant for architecture, construction, real estate, and facilities workflows that already center around floor plan artifacts.
104 use cases in this application