MANAR / منار

Supervised-Autonomy Multisensor Search-and-Rescue System

PROPRIETARY BY DEFAULT · SELECTED COMPONENTS MAY BE OPEN-SOURCED WHEN EXPLICITLY MARKED

MANAR is independently created and owned by Oumar Ibrahim. Unless explicitly licensed otherwise, all source code, engineering material, algorithms, documentation, and project assets in this repository are proprietary.

Component Overview

MANAR Multisensor System Component Overview

Multisensor System Components

Component	Primary Role
Thermal	Person/heat detection
RGB/day	Daytime detection/verification
Low-light/IR	Night visual confirmation
24 GHz FMCW	Presence, range, motion, breathing
Speaker + mic	Prompt, listen, direction finding
Passive RF	Detect/correlate device emissions
Amber beacon	360° visual alert
White strobe	Directional visual guidance
Downward spotlight	Close-range illumination
Heliograph mirrors	Passive daylight signaling
Smoke marker	Location/wind marking

Repository Map

Directory / File	Description
core/	Deterministic C++ control system & flight state engine
perception/vision/	Visual target detection, models (YOLO11n / D-FINE-N), and benchmark suite
perception/mamba/	Mamba SSM temporal multi-sensor fusion research
gui/	Operator Ground Control Station (GCS) staging
landing-page/	Project website
docs/	Technical paper, operator manual, and reference library
assets/	Shared branding, diagrams, and video evaluation datasets
SECURITY.md	Security policy
LICENSE.md	Licensing terms

Key Architectural Decisions

Implemented & Benchmarked Subsystems

Subsystem / Decision	Validation & Rationale
YOLO11n Edge Detector	Empirically selected via multi-spectral benchmark (~29.2 ms CPU latency, ~1.73× faster than D-FINE-N, 0 false alarms on clutter, ~2.6M params) for real-time edge person detection.
Greedy Route Heuristic	Fast, deterministic $O(n^2)$ waypoint sequencing for multi-sector search planning with plan locking and operator override privilege.
Deterministic C++ State Core	State-locked 10 Hz flight state machine, structured JSON command arbitration, and battery-reserve Return-to-Launch (RTL) preemption.

Planned & Investigated Architecture

Area / Concept	Current Scope & Rationale
Mamba SSM Temporal Fusion	Investigating linear $O(L)$ selective state spaces for real-time temporal verification across multi-sensor feature windows as a candidate escalation gate.
1D CNN Feature Projection	Planned feature compressor to align 2D visual candidates into compact 1D embeddings alongside acoustic, radar, and RF anomaly streams.
Two-Level Passive RF	Proposed architecture: Level 1 CFAR anomaly triggering + Level 2 signal derivative trend tracking ($\Delta S_k$).
Multi-Tiered Telemetry	Planned adaptive transmission rate policy (1–2 Hz baseline, 5 Hz proximity, on-event escalation) to prevent RF link saturation.

---

Multimodal Sensing & Target Verification

Synchronized Day RGB, Low-Light IR, and Long-Wave Thermal (LWIR) target verification across diurnal and environmental extremes.

---

Fig. 2. Conceptual 10-stage search, verification, and rescue lifecycle.

MANAR coordinates flight execution through a modular navigation hierarchy and deterministic mission logic.

Fig. 2. Conceptual 10-stage search, verification, and rescue lifecycle.

---

Perception & Sensor-Fusion Pipeline

Synchronized feature vectors $\mathbf{x}_t \in \mathbb{R}^D$ sampled at $10\text{ Hz}$ across $20\text{--}50$ time steps ($2\text{--}5\text{ s}$ window) are evaluated by a Mamba State Space Model (SSM). Mamba acts strictly as an alert escalation filter (Reject candidate, Continue verification, or Alert operator), while final rescue determination is reserved exclusively for the human operator.

Fig. 5. Planned multisensor perception and Mamba hover verification pipeline.

---

Fig. 3. Spatial geometry: Lawnmower sweep coverage pattern (left) and sequential greedy route progression with RTL return leg (right).

When operators specify multiple search sectors, MANAR executes an $O(n^2)$ greedy nearest-neighbor route optimizer to minimize travel transit distance, combined with systematic lawnmower sweeps.

Fig. 3. Spatial geometry: Lawnmower sweep coverage pattern (left) and sequential greedy route progression with RTL return leg (right).

---

System Demonstrations

Current status: Hardware component sizing, React Operator GUI, and downstream multisensor fusion interface.

Autonomous Flight Core & Lawnmower Search

YOLO11n Real-Time Target Detection

---

Try It Yourself (Quick Start)

Requirements: C++17 compliant compiler (g++ / MinGW-w64, Clang, or MSVC) on Windows 10/11, Linux, or macOS.

1. Build the Core Applications

Open PowerShell or terminal in the repository root:

cd core
mkdir -p build

g++ -std=c++17 apps/control.cpp system/shared.cpp system/flight.cpp system/components.cpp system/drone.cpp system/mission.cpp system/route_optimizer.cpp -I. -Ithird_party -o build/control.exe
g++ -std=c++17 apps/terminal.cpp -I. -Ithird_party -o build/terminal.exe
g++ -std=c++17 apps/setup.cpp -I. -Ithird_party -o build/setup.exe

2. Launch the Control Core & Operator Terminal

In Terminal 1 (Flight Control Engine):

cd core
.\build\control.exe

In Terminal 2 (Interactive Operator Command Station):

cd core
.\build\terminal.exe

Tip

For a full mission simulation walkthrough (setting search waypoints, greedy route optimization, sensor toggles, and RTL failsafes) as well as pre-flight parameter configuration (setup.exe), see the Operator Guide.

---

Project Milestones

• System Architecture & Safety Specification: Conceptual 10-stage lifecycle, SWaP constraints, safety invariants, and modular repo hierarchy.
• Deterministic C++ Control Core (core/): 10 Hz flight state machine, greedy route optimizer, failsafe RTL preemption, JSON IPC, and interactive terminal.
• Edge Visual Perception & Benchmark (perception/vision/): Multi-spectral SAR video benchmark, locked YOLO11n ONNX inference engine (~29 ms CPU latency).
• Project Presentation & Website (landing-page/): Responsive project landing page, system diagrams, and public repository showcase.
• Temporal Multi-Sensor Fusion (perception/mamba/): Mamba SSM sequence modeling across synchronized visual, radar, RF, and acoustic streams.
• Modern Operator Ground Station (gui/): Real-time React / TypeScript web interface with geospatial mapping and live telemetry streaming.
• Airframe Engineering & Hardware Validation: Component mass/power budgets, 3D CAD modeling, and propulsion validation for 1-hour active search.
• Academic Publication (docs/paper/): LaTeX technical paper, benchmark analysis, and formal dissemination.

---

Ownership and License

Copyright © 2026 Oumar Ibrahim. All rights reserved.

Unless explicitly stated otherwise, all materials in this repository are proprietary and governed by the MANAR Proprietary Software and Materials License.

Selected files, components, or directories may be released under separate open-source licenses. Any such license applies only to the material explicitly identified as being covered by it.

See the LICENSE and any applicable file or directory license notices for the complete terms.