ABOUT THE DIN

Sub-Canopy Structure Extraction and Confidence Scoring for Synthetic Forest Environments

Negentropic

About the Company

Negentropic is a Sydney-based simulation company solving the terrain bottleneck for physical AI. Its Phantom Sense Engine converts satellite and airborne LiDAR into navigable, physics-accurate virtual worlds where every element carries a confidence score traceable to its source data. Negentropic is a named partner in a $10M CRC-P Round 17 project led by Spiral Blue, with Macquarie University, Space Machines Company, LatConnect 60 and SatRevolution. The company is supported by the AWS Space Accelerator program and has delivered simulation environments for Defence under the Defence SPP, supported by Boeing and RAAF Base Amberley personnel.

Project Objective

Australian and regional operating environments are heavily vegetated, yet most synthetic environments model forest as a canopy surface with empty space beneath. That gap governs occlusion, sensor detection and engagement geometry across mission rehearsal, ISR and guided weapons — closing it from Australian-held data without foreign tooling is a sovereign requirement. This internship sits inside a live $10M CRC-P and tasks the intern with extending the Phantom Sense Engine toward sub-canopy structure recovery: improving ground inference beneath dense canopy, extracting understorey density, trunk positions and vertical stratification, and replacing placeholder reliability scores with measurable per-element confidence values.

Project Tasks

• Weeks 1–5 (Ground inference): Onboard to Phantom Sense pipeline; characterise failure modes of current ground interpolation on sparse ground returns; prototype improved ground surface recovery; test across staged forested datasets. Output: documented failure analysis, prioritised backlog, working ground inference prototype.
• Weeks 5–9 (Sub-canopy structure): Develop understorey density, trunk position and vertical stratification extraction; benchmark against the current classifier; write test coverage; integrate into the pipeline. Output: sub-canopy extraction module with automated test suite and comparative benchmark.
• Weeks 10–12 (Confidence scoring and handover): Define and implement per-element uncertainty for sub-canopy features; quantify fidelity improvement; write method documentation; deliver handover pack and final presentation to Negentropic and CRC-P partners.

Intern Skills

• Python and/or C# with Git; comfort with noisy, incomplete data matters more than any specific tool.
• Familiarity with several of: point cloud processing (laspy, PDAL); raster and geospatial data (rasterio, GDAL); spatial statistics or clustering; 3D graphics or game engines (Unity, Unreal, Three.js); automated testing. Prior vegetation/forestry LiDAR experience is a strong advantage.
• Postgraduate study in computer science, software engineering, geospatial science, surveying, forestry or remote sensing.
• Australian citizenship required 

How to apply

Eligible students (Australian citizens) apply via the online application form and submit a CV and motivation letter by 6 September 2026.