🛰️ Streaming Airborne LiDAR Simulator (Rotating Multi-Beam)

A Unity-based simulator for airborne rotating multi-beam LiDAR with real-time point cloud generation and on-the-fly scan-quality assessment.

---

🧭 Overview

This project simulates a VLP-16-equivalent rotating multi-beam LiDAR mounted on a low-altitude UAV flying an automatic S-pattern (boustrophedon) survey.

Unlike offline simulators that buffer everything and save after the mission, this one streams the point cloud and writes quality metrics row-by-row during flight, with a constant memory footprint and sub-millisecond per-sweep assessment latency.

Designed for UAV-LiDAR algorithm prototyping, pre-flight validation, and real-time data-quality feedback.

---

🔭 1. Simulation Principle

• Rotating multi-beam emission: 16 beams over a 30° along-track fan, rotated about the flight axis at 10 Hz; only the ground-facing half-revolution (|ω| ≤ 75° from nadir) fires.
• Native pulse rate: 50 Hz FixedUpdate × 360 firings/step = 18 kHz, i.e. ~28,800 pts/rev, matching the real VLP-16 (ω/f = 0.2°/firing).
• Sub-stepped raycasting: platform pose is linearly interpolated within each physics step so fast rotation is sampled correctly via Physics.Raycast.
• Platform-motion model: along-track acceleration and Gaussian vibration (cross-track / vertical) can be enabled to study their effect on point-cloud quality.

---

🧱 2. Core Components

Script	Role	Key Methods
SPatternScanFlightPhased.cs	UAV flight control, S-pattern trajectory, beam geometry, raycasting, point-cloud export	PerformScanSubstepped(), ExecuteStraightFlight(), ExecuteTurning()
StreamingQualityAssessment.cs	Streaming sweep-level quality metrics (ring-buffer, incremental, constant memory)	OnFiring(), CommitCurrentSweep(), ComputeSweepCCI()

---

🌲 3. Scene Setup (RAMI / LESS terrain)

Scenes are built from RAMI / LESS forest reconstructions (e.g. HET09_JBS_SUM) by Qi et al. from https://lessrt.org/res_html/rami.html and prepared for raycast-only scanning. Both are Editor tools.

Script	Role	Entry point
RAMISceneBuilder.cs	Instantiate OBJ objects from objects.txt + instances.txt, convert LESS (X,Y,Z, Z-up) → Unity (X,Z,Y, Y-up), auto-align ground to y=0, optional GPU instancing	Menu RAMI ▸ Build Scene from LESS
RamiSceneTiler.cs	Detect a single patch's XZ extent and tile the prefab into a large tilesAlongX × tilesAlongZ scene	Inspector buttons Auto Detect Extent / Build Tiles

Typical workflow:

1. Put the RAMI parameter files under Assets/RAMI/<scene>/Parameters/.
2. RAMISceneBuilder → Build Scene from LESS to instantiate the patch.
3. Convert to Colliders — removes MeshRenderers and adds MeshColliders so Physics.Raycast hits the geometry (rendering not needed for scanning; also avoids the PhysX Fast-Midphase crash on >2²¹-triangle meshes).
4. Save the result as a prefab, assign it to RamiSceneTiler, Auto Detect Extent → Build Tiles to cover the full survey area.

---

📐 4. Parameters

LiDAR (SPatternScanFlightPhased)

Parameter	Description	Default
lines	Beam/channel count	16
verticalFOV	Along-track fan angle Θ (°)	30
maxDetectionDistance	Max range (m)	100
rotationFrequency	Scanner rotation (Hz)	10
maxScanAngleFromNadir	Ground-facing half-angle (°)	75
firingsPerFixedUpdate	Firings per physics step	360

Flight

Parameter	Description	Default
targetHeight	AGL altitude (m)	50
cruiseSpeed	Cruise speed (m/s)	5
swathWidth / swathLength	Line spacing / length (m)	20 / 400
numberOfSwaths	Number of survey lines	5
turnRadius / maxBankAngle	Turn radius (m) / bank limit (°)	30 / 30
cruiseAcceleration	Along-track accel a (m/s², 0 = off)	0
vibrationSigma	Vibration σ_v (m/s, 0 = off)	0

Quality Metrics (StreamingQualityAssessment)

Parameter	Description	Default
WindowSize	Sliding window W (sweeps)	50
PDVRThreshold	Point density variation alert	0.10
ISCThreshold	Inter-sweep spacing alert	0.15
CCIThreshold	Coverage completeness alert	0.60
CCIBinCount	Cross-track bins Q	4

---

📊 5. Quality Metrics

Three complementary metrics are computed incrementally, one value per ground-facing sweep:

Metric	Meaning	Flag when
PDVR	Point Density Variation Rate between consecutive sweeps	> 0.10
ISC	Inter-Sweep spacing Consistency (CV of sweep spacing)	> 0.15
CCI	Coverage Completeness Index (actual / expected per cross-track bin × uniformity)	< 0.60

Controlled single-effect experiments

To isolate how each platform motion affects the metrics, three minimal scanners reuse the exact same scan kernel as the full S-pattern but replace the trajectory with one clean motion:

Script	Motion	Tests
CruiseScanner.cs	Constant-velocity straight line	Steady-state baseline (PDVR ≈ 0, low ISC, CCI ≈ const)
AccelScanner.cs	Constant along-track acceleration	PDVR response to density drift ((1/D)(dD/dt) = −a/v)
BankingScanner.cs	Straight → single 180° coordinated turn → straight	CCI drop / ISC rise during banked turn

Each writes the same metric CSV with a Phase column (Cruise / Accel / Straight / Turning) so the effect is directly traceable in the output.

---

💾 6. Data Output

All files are written to /PointClouds/<sessionId>/ (sessionId = yyyyMMdd_HHmmss).

File	Content
QualityMetrics_<id>.csv	Per-sweep ScanIndex,Timestamp,PDVR,ISC,CCI,Warning,Phase,Pi,RhoI
QualitySummary_<id>.txt	Aggregate mean / max / threshold-exceedance summary
Performance_<id>.csv	Per-step Step,Timestamp,ScanMs,MetricMs,MemMB (latency & memory)
SPattern_PointCloud_<id>.txt	x y z range intensity scan_angle (if savePointCloud)
SPattern_RangeMatrix_<id>.txt	Per-scan range rows (if saveRangeMatrix)
SPattern_Trajectory_<id>.txt	Per-step platform pose s_t (if saveTrajectory)

For performance/time-series runs, set savePointCloud / saveRangeMatrix / enablePointDisplay to false to keep memory constant.

---

🚀 7. Usage

1. Open the scene in Unity 2022.3.62f and build the terrain (see Section 3).
2. Attach SPatternScanFlightPhased to the UAV object and StreamingQualityAssessment to any object, then assign the latter to the qualityAssessor field.
3. Set LiDAR / flight / metric parameters in the Inspector.
4. Press Play — scanning starts automatically; metrics stream live and a quality panel renders the point cloud.
5. The flight stops at the last swath; all CSV/TXT outputs are saved automatically.

For the controlled experiments, use CruiseScanner / AccelScanner / BankingScanner instead of SPatternScanFlightPhased (same setup, single motion).

---

📈 8. Visualization & Analysis

• Import *_PointCloud_*.txt into CloudCompare, Open3D, or MATLAB.
• Plot QualityMetrics_*.csv to inspect PDVR / ISC / CCI over the flight.
• Use Performance_*.csv to verify constant memory and sub-ms metric latency.

---

🧩 9. Future Work

• Reflectivity / noise modeling
• PLY / PCD export
• Additional flight patterns and terrain loaders

---

📜 License

Released under the MIT License. Free for research and educational use with attribution.

---

🇨🇳 中文简介

基于 Unity 的机载旋转多线束 LiDAR 仿真器(VLP-16 等效),无人机自动执行 S 型航线扫描, 地形来自 RAMI / LESS 森林重建场景。飞行过程中实时流式生成点云并逐行计算扫描质量指标 (PDVR / ISC / CCI),内存占用恒定,单 sweep 评估延迟亚毫秒。另附 Accel / Banking/ Cruise 三个受控实验脚本,分别隔离匀速、加速、转弯对点云质量的影响。可用于 UAV-LiDAR 算法原型验证、 飞行前测试与实时质量反馈。