- SLAM-MIPSLAM // W7
- Python
- ANIMA
SLAM-MIPSLAM
Crisp Maps at Any Scale, Without the Aliasing Lies
Standard 3D Gaussian Splatting SLAM renders beautifully at the training resolution but produces aliased, geometrically incorrect outputs when rendered at different scales — a critical failure mode when an operator zooms into a map to inspect a chokepoint, or when a downstream planner requests a lower-resolution occupancy grid. Aliasing is not cosmetic; it shifts apparent obstacle positions. SLAM-MIPSLAM (codename: Susanoo) implements MipSLAM (arXiv:2603.06989), introducing EAA (Exact Anti-Aliased) rendering into the Gaussian SLAM pipeline alongside SA-PGO (Scale-Aware Pose Graph Optimization).
MODULE STATUS: PRODUCTIONAnti-Aliasing
PROD
- DIVISION
- GENERIC
- WAVE
- W7
- DOMAIN
- GENERAL
- WAVE 7 // ANIMA SUITE
- FOUNDATION — SLAM-MIPSLAM
THE PROBLEM WE SOLVE
Standard 3D Gaussian Splatting SLAM renders beautifully at the training resolution but produces aliased, geometrically incorrect outputs when rendered at different scales — a critical failure mode when an operator zooms into a map to inspect a chokepoint, or when a downstream planner requests a lower-resolution occupancy grid. Aliasing is not cosmetic; it shifts apparent obstacle positions.
Operators and downstream planners can trust map outputs at any zoom level — from full-site overview to centimeter-precision inspection — with geometric fidelity guaranteed by construction, not approximation.
WHAT SLAM-MIPSLAM DELIVERS
SLAM-MIPSLAM (codename: Susanoo) implements MipSLAM (arXiv:2603.06989), introducing EAA (Exact Anti-Aliased) rendering into the Gaussian SLAM pipeline alongside SA-PGO (Scale-Aware Pose Graph Optimization). EAA analytically filters Gaussian contributions at each target resolution, eliminating the frequency aliasing that corrupts standard 3DGS at off-training scales.
CAPABILITIES
- SLAM-MIPSLAM (codename: Susanoo) implements MipSLAM (arXiv:2603.06989), introducing EAA (Exact Anti-Aliased) rendering into the Gaussian SLAM pipeline alongside SA-PGO (Scale-Aware Pose Graph Optimization)
- EAA analytically filters Gaussian contributions at each target resolution, eliminating the frequency aliasing that corrupts standard 3DGS at off-training scales
- SA-PGO ensures pose graph optimization accounts for scale-dependent geometric consistency
- The module supports multi-resolution evaluation and exports trajectory in standard formats.
WHY THIS IS HARD
Building SLAM-MIPSLAM requires solving multiple coupled problems:
- 01Standard 3D Gaussian Splatting SLAM renders beautifully at the training resolution but produces aliased, geometrically incorrect outputs when rendered at different scales — a critical failure mode when an operator zooms into a map to inspect a chokepoint, or when a downstream planner requests a lower-resolution occupancy grid
- 02Aliasing is not cosmetic; it shifts apparent obstacle positions.
- 03SLAM-MIPSLAM (codename: Susanoo) implements MipSLAM (arXiv:2603.06989), introducing EAA (Exact Anti-Aliased) rendering into the Gaussian SLAM pipeline alongside SA-PGO (Scale-Aware Pose Graph Optimization)
- 04EAA analytically filters Gaussian contributions at each target resolution, eliminating the frequency aliasing that corrupts standard 3DGS at off-training scales
SLAM-MIPSLAM solves these through careful architecture design and rigorous validation.
PROOF, NOT PROMISES
Key performance metrics:
| METRIC | VALUE |
|---|---|
| Anti-Aliasing | EAA rendering — mathematically exact filtering at arbitrary target resolution |
| Backend Flexibility | CUDA + macOS MLX/MPS/CPU via ANIMA_BACKEND=auto — runs on any hardware tier |
| Multi-Resolution Eval | Built-in evaluation harness across resolution scales — quantitative fidelity at every zoom |
| Defense Angle | Scale-trustworthy mapping for mission planning dashboards, precision navigation, and sensor fusion pipelines |
WHAT'S BUILT TODAY
| COMPONENT | STATUS | NOTES |
|---|---|---|
| Anti-Aliasing | COMPLETE | EAA rendering — mathematically exact filtering at arbitrary target resolution |
| Backend Flexibility | COMPLETE | CUDA + macOS MLX/MPS/CPU via ANIMA_BACKEND=auto — runs on any hardware tier |
| Multi-Resolution Eval | IN PROGRESS | Built-in evaluation harness across resolution scales — quantitative fidelity at every zoom |
| Defense Angle | IN PROGRESS | Scale-trustworthy mapping for mission planning dashboards, precision navigation, and sensor fusion pipelines |
WHERE SLAM-MIPSLAM DEPLOYS
- APP_01
AUTONOMOUS SYSTEMS
Operators and downstream planners can trust map outputs at any zoom level — from full-site overview to centimeter-precision inspection — with geometric fidelity guaranteed by construction, not approximation.
- APP_02
RESEARCH LABS
SLAM-MIPSLAM (codename: Susanoo) implements MipSLAM (arXiv:2603.06989), introducing EAA (Exact Anti-Aliased) rendering into the Gaussian SLAM pipeline alongside SA-PGO (Scale-Aware Pose Graph Optimization).
- APP_03
EDGE COMPUTING
Alias-free Gaussian SLAM that renders accurate maps at multiple resolutions — because a map that looks sharp but is geometrically wrong is a safety failure, not a metric.