Active R&D · Subsurface Inspection
Project Turtle
Pocket-LiDAR subsurface inspection — iPhone Pro capture, Jetson Thor fusion, browser-native 3D walk-through. Active R&D.
The problem we're solving
Handheld capture. Engineering-grade output.
Subsurface infrastructure — pipes, culverts, manholes, and buried conduits — is expensive to inspect, difficult to document, and nearly impossible to share in three dimensions with engineering fidelity. Conventional inspection workflows depend on dedicated hardware, separate data processing pipelines, and deliverables that live in file formats most stakeholders cannot open.
Project Turtle collapses that stack. An iPhone Pro running AK-mee's capture application records LiDAR depth, camera frames, and ARKit pose simultaneously. That data travels to NVIDIA Jetson Thor for sensor fusion and reconstruction — producing a spatially-accurate point cloud, a mesh, and a 3D Gaussian Splat that any project-authorized stakeholder can walk through in a browser, without installing anything.
The pipeline
Capture. Fuse. Splat. View.
- Capture — iPhone Pro (ARKit + LiDAR Scanner). The iOS application records synchronized LiDAR depth frames, EO imagery, and ARKit pose data. Bluetooth remote support enables hands-free operation in confined access environments. Raw scan data is packaged on-device and transferred to the backend over the local network.
- Sensor fusion — Jetson Thor (TSDF + ICP). Depth frames and camera imagery are fused via Truncated Signed Distance Function (TSDF) volumetric integration on the Jetson's CUDA pipeline. Multi-scan registration uses Iterative Closest Point (ICP) with GPS fallback to align multiple capture runs into a single coherent coordinate space.
- Reconstruction — Open3D + gsplat. The fused volume is reconstructed into a dense point cloud and mesh via Open3D. A 3D Gaussian Splat is trained against the captured imagery using gsplat, producing a visually accurate rendering asset for the browser viewer.
- Browser viewer — project-member access. Reconstructed assets are served via a FastAPI backend and rendered in a browser-native 3D viewer. No plugin, no desktop app, no format conversion on the client side. Access is gated to named project members via Cloudflare Access SSO.
Current milestone
Where we are today.
- Sensor Fusion v2 (TSDF) shipped 2026-05-30. Full volumetric TSDF integration pipeline on CUDA; replaces the earlier BPA-only mesh path.
- Multi-scan registration (ICP + GPS fallback) shipped. Multiple capture runs align automatically into a single coordinate frame.
- iOS Build 22 with Bluetooth remote support shipped. Hands-free trigger for confined-space capture scenarios.
- Field testing 2026-05-31. End-to-end pipeline validation against live subsurface geometry.
The viewer
Walk through it in your browser.
The Project Turtle viewer renders the latest reconstruction assets — point cloud, mesh, and 3D Gaussian Splat — directly in the browser. No software to install. Access is restricted to named project members via Cloudflare Access SSO.
Sign in with your akmee.net account — viewer access is project-members-only via Cloudflare Access.
AK-mee Engineering · Subsurface & Sensor Fusion · AK-mee™ · Updated 2026-05-30