SLAM and spatial intelligence software technology and investment research

Technical bottleneck

• Bottleneck type: Algorithm robustness / Computational efficiency
• Technical constraint: SLAM is a chicken-and-egg problem — localization requires a map, mapping requires localization. Visual SLAM at >30 fps requires tracking 200-500 features per frame with <5ms feature extraction and matching (ORB, SuperPoint); loop closure detection (recognizing you've returned to a previously-visited place) requires a bag-of-words database lookup across millions of images — false loop closures cause catastrophic map corruption; lifelong SLAM (operating in the same environment for months/years) must handle lighting changes, moved furniture, seasonal outdoor changes — the map must adapt without forgetting
• Economic constraint: SLAM is mostly open-source (ORB-SLAM3, OpenVSLAM, Kimera, RTAB-Map) and embedded in larger robot platforms (NVIDIA Isaac, ROS 2 navigation stack); few pure-play SLAM companies exist; exposure comes through: (1) robot OEMs that build proprietary SLAM (iRobot, DJI), (2) chip vendors that optimize SLAM for their silicon (Qualcomm RB6, NVIDIA Jetson), and (3) SLAM-as-a-service for specific verticals (AR/VR spatial anchors)

Adoption

• Driver: AR/VR inside-out tracking (Meta Quest, Apple Vision Pro) requiring robust vSLAM on mobile SoCs; warehouse AMR visual SLAM replacing magnetic tape/QR codes; drone GPS-denied navigation; autonomous vacuum market growth
• Blocker: Open-source SLAM (ORB-SLAM3, RTAB-Map) adequate for most applications — limited commercial TAM for proprietary SLAM software; LiDAR-based SLAM simpler and more robust than visual — reducing vSLAM's advantage; SLAM embedded in robot platforms as a commodity feature

Public companies exposed

NVDA (NVIDIA Isaac — Isaac VSLAM

cuVSLAM GPU-accelerated library)

QCOM (Qualcomm — Snapdragon Spaces

vSLAM optimised for RB6)

ISRG (Intuitive Surgical — proprietary SLAM for da Vinci SP)

IRBT (iRobot — vSLAM for Roomba

proprietary)

ADSK (Autodesk — does some reality capture

not core SLAM)

META (Meta — Quest inside-out vSLAM

captive)

AAPL (Apple — Vision Pro vSLAM

ARKit

captive)

Validation signals

SLAM capability becoming a standard feature in robot platforms (Jetson, RB6); AR/VR headset shipments driving vSLAM silicon optimizations; warehouse AMR deployments specifying visual SLAM for infrastructure-free navigation

Invalidation signals

LiDAR cost reduction making vSLAM's cost advantage irrelevant; open-source SLAM closing the robustness gap with proprietary; UWB/5G indoor positioning providing infrastructure-based localization at lower compute