Al-Powered Underground Mine Safety, Monitoring and Rescue System.

1. Executive Snapshot

Summary

Underground coal mining operations in Jharkhand face extreme environmental hazards including flammable methane gas explosions ($CH_4$), toxic carbon monoxide ($CO$) accumulation, oxygen deficiency, flooding, and roof/gallery collapses [PS, KNOWN]. When disasters occur, traditional rescue efforts are severely constrained by zero visibility, loss of wired/wireless communication, and high risk of secondary explosions or collapses that threaten rescuers' lives [PS, KNOWN]. The proposed system is an integrated AI-Powered Mine Safety and Rescue System featuring a rugged, tracked ground rover equipped with multi-gas sensors, thermal and night-vision vision systems, LiDAR-based 3D mapping (SLAM), an automated droppable wireless mesh node deployment mechanism, and a tethered mini-drone [INFERENCE, ASSUMPTION]. Powered by an NVIDIA Jetson edge AI processor and microcontrollers running ROS 2 [ASSUMPTION], the rover navigates GPS-denied galleries, detects environmental hazards, locates trapped miners via thermal YOLO vision models, and relays real-time video, gas telemetry, and structural 3D maps back to a surface control station via a multi-hop wireless mesh [INFERENCE].

System Metadata

Fundamental System Objective

To provide surface rescue teams with immediate, high-fidelity, real-time situational awareness—including toxic gas concentrations, thermal human detection, structural 3D mapping, and low-latency video—from hazardous, GPS-denied underground mine galleries without risking human lives [PS, INFERENCE].

2. Problem Deconstruction

The explicit problem statement provided by the Government of Jharkhand is broken down into atomic requirements below:

• Requirement (a): Multi-Gas and Environmental Condition Sensing

  • Source Text: "...equipped with gas sensors... environmental monitoring sensors... detect toxic gases, monitor temperature and humidity..." [PS]
  • Interpretation: The vehicle must continuously sample atmospheric composition ($CH_4$, $CO$, $CO_2$, $O_2$) along with ambient ambient temperature and relative humidity in real time [INFERENCE].
  • Implementation Implication: Sensors must feed into an analog/digital microcontroller (ESP32 / STM32) with calibrated thresholds, fast response times ($T_{90} < 15\text{s}$), and local visual/audible status warnings [INFERENCE].
  • Tag: [PS]

• Requirement (b): Thermal & Night-Vision Visual Sensing

  • Source Text: "...thermal and night-vision cameras... transmitting live video and thermal imaging data..." [PS]
  • Interpretation: Due to complete darkness, dust, and thick smoke in mine shafts, standard RGB cameras fail [KNOWN]. The vehicle requires long-wave infrared (LWIR) thermal imaging and low-light IR-illuminated cameras [INFERENCE].
  • Implementation Implication: Dual-camera payload (LWIR Thermal camera e.g., FLIR Lepton 3.5 / AMG8833 + IR-cut night-vision sensor) processed on an edge AI board [ASSUMPTION].
  • Tag: [PS]

• Requirement (c): Structural Condition & Hazard Identification

  • Source Text: "...identify potential hazards... monitor... structural conditions... explore inaccessible areas..." [PS]
  • Interpretation: The system must identify gallery blockages, roof instabilities, water accumulation/flooding, and physical obstacles in narrow underground cross-cuts [INFERENCE].
  • Implementation Implication: Integration of 2D/3D LiDAR or solid-state depth cameras to perform Simultaneous Localization and Mapping (SLAM) and volumetric obstacle evaluation in real time [INFERENCE].
  • Tag: [PS]

• Requirement (d): Trapped Worker Detection & Localization

  • Source Text: "...assist in locating trapped workers during emergencies..." [PS]
  • Interpretation: The system must autonomously detect human presence in low-light, smoky environments and estimate their spatial coordinates relative to the entry portal [INFERENCE].
  • Implementation Implication: Edge AI inference (YOLOv8/v10 fine-tuned on thermal/infrared human signature datasets) combined with SLAM spatial coordinate mapping [RESEARCHED, INFERENCE].
  • Tag: [PS]

• Requirement (e): Underground-to-Surface Wireless Data Transmission

  • Source Text: "...wireless communication modules to provide real-time information... transmitting live video and thermal imaging data to a surface control station." [PS]
  • Interpretation: Rock strata, coal pillars, and metal machinery cause massive radio frequency (RF) attenuation, rendering direct line-of-sight wireless useless past tunnel turns [KNOWN].
  • Implementation Implication: Hybrid multi-hop communication architecture utilizing Sub-1GHz (LoRa Mesh) for telemetry/alerts and 2.4GHz/5.8GHz Wi-Fi Mesh with automated droppable repeater nodes for compressed video/maps [RESEARCHED, INFERENCE].
  • Tag: [PS]

• Requirement (f): Unmanned Hardware Platform (Rover / Compact Drone)

  • Source Text: "...consisting of a rugged ground rover or a compact aerial drone capable of operating in hazardous underground mining environments." [PS]
  • Interpretation: The vehicle chassis must cross uneven rubble, mud, water, and steep inclines ($>25^\circ$) typical of coal seam roadways [KNOWN].
  • Implementation Implication: High-torque, differential-steer tracked chassis with IP67 waterproofing, flameproof/sealed battery enclosures, and optional micro-tethered drone for high-roof gallery inspection [INFERENCE, ASSUMPTION].
  • Tag: [PS]

3. Stakeholders & User Needs

4. Requirements Traceability

5. Constraints & Environment Realities

• Explosive Atmospheric Hazard (Gassy Mines): Underground coal mines contain explosive Methane ($CH_4$) gas (explosive limit: 5%–15% in air) and combustible coal dust [KNOWN]. Any electrical spark, hot motor casing ($>135^\circ\text{C}$), or open switch can trigger a catastrophic explosion [KNOWN]. Equipment in production must adhere to DGMS flameproof (FLP / Ex-d under IS/IEC 60079-1) or intrinsically safe (IS / Ex-ia under IS/IEC 60079-11) standards [RESEARCHED].
• Extreme RF Attenuation & Multipath Scattering: Radio waves at 2.4GHz and 5.8GHz attenuate rapidly around solid coal seams and rock pillars (signal loss $>30\text{ dB}$ around sharp $90^\circ$ bends) [KNOWN]. Through-the-earth (TTE) propagation is infeasible for broadband video; non-line-of-sight (NLOS) multi-hop relaying is strictly required [RESEARCHED, INFERENCE].
• Complete Darkness, Thick Dust, and Smoke: Airborne coal dust and dense smoke render conventional optical cameras blind [KNOWN]. Active infrared illumination and long-wavelength thermal imaging are non-negotiable [INFERENCE].
• Severe Terrain Obstacles: Flooded galleries ($10–20\text{ cm}$ water depth), loose rock debris, mud, steep gradients up to $30^\circ$, and narrow cross-cuts ($<1\text{ meter}$) prevent standard wheeled vehicles from progressing [KNOWN, INFERENCE].
• GPS-Denied Environment: Satellite positioning signals cannot penetrate rock strata [KNOWN]. All localization must rely strictly on dead-reckoning, wheel encoders, IMU sensor fusion, and LiDAR/Visual SLAM [INFERENCE].
• Field Operator Constraints: Mine rescue teams operate in high-stress, low-visibility conditions wearing thick gloves and rebreathers [KNOWN]. Dashboard interfaces must be intuitive, high-contrast, touch-friendly, and require minimal manual configuration [INFERENCE].

6. Research Questions

1. Gas Sensing Reliability: What is the drift and response time of low-cost electrochemical and NDIR methane sensors in high-humidity ($>90%$), dust-heavy underground coal mine atmospheres? [NEEDS VALIDATION]
2. RF Propagation & Mesh Drop Frequency: What is the maximum effective transmission distance for Sub-GHz LoRa (868/915 MHz) and 2.4GHz Wi-Fi mesh through jagged, damp underground coal galleries before a repeater node must be dropped? [RESEARCHED, NEEDS VALIDATION]
3. Thermal YOLO Accuracy: How effectively does a lightweight YOLOv8 model trained on thermal infrared imagery detect trapped workers covered in coal dust or partially submerged under debris? [RESEARCHED]
4. Battery Thermal Limits: What battery chemistries ($LiFePO_4$ vs $LiPo$) provide the required energy density for a 2-hour rover mission while meeting DGMS surface temperature safety limits ($<135^\circ\text{C}$) without active cooling vents? [RESEARCHED, INFERENCE]
5. Aerial vs Ground Traversal: Is a hybrid tethered mini-drone feasible for high-roof cavity inspection without creating rotor wash that suspends explosive coal dust? [INFERENCE, NEEDS VALIDATION]

7. Data Requirements & Sources

8. Existing Solutions & Gap Analysis

9. Candidate Solution Approaches

Approach 1: Heavy Tethered Ground Rover with Analog Video Relay

• Mechanism: A heavy, multi-wheeled rover connected to the surface base station via a $300\text{-meter}$ reinforced fiber-optic/copper tether cable. Analog video and serial telemetry are transmitted directly through the cable [INFERENCE].
• Required Data: Analog video stream, raw analog sensor voltages, encoder ticks [INFERENCE].
• Complexity & Cost: Low complexity; moderate cost ($\sim\text{₹}80,000\text{--}\text{₹}1,200,000$) [INFERENCE].
• Pros & Cons: Eliminates RF signal attenuation; guaranteed high-bandwidth video [INFERENCE]. However, the tether frequently snags around sharp rock outcrops, gets crushed under falling debris, severely restricts mobility, and adds massive weight [INFERENCE].
• Hackathon Feasibility: High | Production Feasibility: Low (Tether snagging is unacceptable in collapse zones) [INFERENCE].

Approach 2: Autonomous Aerial Drone Swarm with Wi-Fi Mesh

• Mechanism: A fleet of 2–3 compact quadcopters equipped with optical flow, thermal cameras, and micro Wi-Fi mesh modules flying autonomously into mine shafts [INFERENCE].
• Required Data: High-rate IMU ($>200\text{Hz}$), optical flow vector streams, thermal video, 3D point cloud [INFERENCE].
• Complexity & Cost: High complexity; high cost ($\sim\text{₹}1,500,000\text{--}\text{₹}3,000,000$) [INFERENCE].
• Pros & Cons: Rapid flight over flooded or blocked gallery floors [INFERENCE]. However, extreme battery drain ($\sim10\text{-minute}$ flight time), high rotor dust turbulence, collision risks in narrow $1\text{-meter}$ cross-cuts, and high power usage make it unreliable for prolonged search operations [INFERENCE].
• Hackathon Feasibility: Low | Production Feasibility: Medium (Flight time constraints) [INFERENCE].

Approach 3 (RECOMMENDED): Modular Tracked Rover + Droppable Mesh Node Deployer + Surface Dashboard (Hybrid Architecture)

• Mechanism: A rugged, tracked ground rover equipped with an onboard NVIDIA Jetson edge AI board, multi-gas array, 2D LiDAR, thermal camera, and an automated servo-actuated droppable mesh node mechanism [INFERENCE]. As the rover drives into the mine gallery, it continuously monitors RF signal strength (RSSI) and automatically drops a small, battery-powered Sub-GHz / Wi-Fi repeater node before turning corners or when signal drops below $-75\text{ dBm}$ [INFERENCE].
• Required Data: LiDAR scans (/scan), thermal image array, RGB video, gas sensor readings ($CH_4, CO, Temp$), mesh node RSSI telemetry [INFERENCE].
• Complexity & Cost: Moderate-High complexity; affordable cost ($\sim\text{₹}60,000\text{--}\text{₹}1,200,000$ prototype) [INFERENCE, ASSUMPTION].
• Pros & Cons: Bypasses RF line-of-sight attenuation via droppable breadcrumbs; high mobility over rubble; $2\text{--}3\text{ hour}$ operation time; safe low-center-of-gravity traversal [INFERENCE].
• Hackathon Feasibility: High | Production Feasibility: High [INFERENCE].

10. Recommended Approach & Why

Selected Solution

Approach 3: Modular Tracked Ground Rover with Automated Droppable Wireless Mesh Relay Nodes, Dual Thermal-Visual Sensing, Edge AI, and Surface Command Dashboard [INFERENCE, ASSUMPTION].

Why Alternatives Were Rejected

• Approach 1 (Tethered Rover): Rejected because physical cable tethers inevitably snag, twist, or get severed by jagged rock falls in collapsed mine galleries, causing complete mission failure [INFERENCE].
• Approach 2 (Pure Aerial Drone Swarm): Rejected due to short battery flight times ($<15\text{ min}$), heavy dust displacement from propeller wash interfering with optical sensors, and high crash risk in tight, unmapped cross-cuts [INFERENCE].

Primary Winning Differentiator

Approach 3 directly solves the single greatest technical bottleneck of underground mining robotics: RF signal death around rock corners [KNOWN]. By dynamically deploying low-cost, battery-powered droppable mesh repeater nodes ("breadcrumbs") [RESEARCHED], the rover creates a self-healing communication pipeline back to the surface control station while maintaining continuous ground mobility [INFERENCE].

11. Technical Architecture

Layered Architecture Diagram

+-----------------------------------------------------------------------------------+
|                            SURFACE BASE STATION LAYER                             |
|  +-----------------------------------------------------------------------------+  |
|  | Surface Command Dashboard (React.js / WebSockets / Chart.js / Leaflet GIS)   |  |
|  | - Live Dual Video (Thermal + IR)  - Real-time Gas Gauges ($CH_4$, CO, $O_2$)  |  |
|  | - 2D/3D SLAM Map & Worker Pins    - Node Mesh Health & RSSI Diagnostics     |  |
|  +-----------------------------------------------------------------------------+  |
|                                       ^                                           |
|                   UDP / WebSockets over Wi-Fi & Serial-LoRa                       |
|                                       v                                           |
+-----------------------------------------------------------------------------------+
|                        WIRELESS COMMUNICATION & MESH LAYER                        |
|  +-----------------------+      +-----------------------+     +----------------+  |
|  | Surface Base Node     |<---->| Droppable Mesh Node 2 |<--->| Droppable Node1|  |
|  | (ESP32 + LoRa/Wi-Fi)  |      | (ESP32 / ESP-NOW)     |     | (Dropped Node) |  |
|  +-----------------------+      +-----------------------+     +----------------+  |
|                                                                       ^           |
|                                                    Wi-Fi / LoRa Link  |           |
|                                                                       v           |
+-----------------------------------------------------------------------------------+
|                          ONBOARD ROVER PROCESSING LAYER                           |
|  +-----------------------------------------------------------------------------+  |
|  | MAIN COMPUTE: NVIDIA Jetson Orin Nano (ROS 2 Humble / Linux Ubuntu)         |  |
|  |  - SLAM Engine (Cartographer / SLAM Toolbox) -> 2D Occupancy Grid Map       |  |
|  |  - Edge AI Vision (YOLOv8n TensorRT) -> Thermal Human Body Detection        |  |
|  |  - Navigation & Teleop Node -> Differential Drive Control                   |  |
|  +-----------------------------------------------------------------------------+  |
|                                       ^ (UART / micro-ROS)                        |
|                                       v                                           |
|  +-----------------------------------------------------------------------------+  |
|  | SENSOR & ACTUATOR CONTROLLER: ESP32 / STM32 Microcontroller                 |  |
|  |  - Gas Sampling Loop ($CH_4, CO, CO_2, O_2, Temp, Humidity$)                 |  |
|  |  - Motor Drive Bridge (BTS7960 PWM Driver -> High-Torque Tracked Motors)     |  |
|  |  - Automated Mesh Deployer (Servo Motor Dispenser Trigger)                 |  |
|  +-----------------------------------------------------------------------------+  |
+-----------------------------------------------------------------------------------+
|                            PHYSICAL HARDWARE LAYER                                |
|  +-----------------------------------------------------------------------------+  |
|  | - RPLiDAR A1M8 (2D LiDAR)     - FLIR Lepton 3.5 / AMG8833 Thermal Array      |  |
|  | - Night Vision IR Camera      - MQ-4 / MQ-7 / MQ-135 / DHT22 Gas Sensors      |  |
|  | - High-Torque Tracked Chassis - 12V LiFePO4 Battery Pack (Sealed Flameproof)   |  |
|  +-----------------------------------------------------------------------------+  |
+-----------------------------------------------------------------------------------+

Data-Flow Narrative

1. Sensing & Local Acquisition: The ESP32 sensor controller samples gas concentrations ($MQ-4, MQ-7, DHT22$) every $100\text{ms}$. If Methane exceeds $1.2%$ or $CO$ exceeds $50\text{ ppm}$, an immediate local interrupt triggers an audible buzzer and high-priority alert flag [INFERENCE].
2. Onboard AI & Mapping Processing: The NVIDIA Jetson Orin Nano receives 2D LiDAR scans over USB (/scan) and runs slam_toolbox to generate a live occupancy grid [RESEARCHED]. Concurrently, thermal imagery from the LWIR sensor is ingested into a TensorRT-optimized YOLOv8n model [RESEARCHED]. When a human thermal signature is detected with confidence $>0.70$, the pipeline projects the bounding box onto the current SLAM map coordinates, generating a "Trapped Miner Marker" [INFERENCE].
3. Automated Breadcrumb Mesh Deployment: As the rover moves deeper into the gallery, the Jetson continuously monitors RSSI from the last connected relay node. If RSSI drops below $-75\text{ dBm}$ or a sharp $90^\circ$ turn is detected via IMU, Jetson commands the ESP32 to actuate a servo dispenser, dropping a battery-powered ESP32 mesh node onto the gallery floor [RESEARCHED, INFERENCE].
4. Data Transmission & Surface Visualization: Compressed telemetry packages (gas levels, pose coordinates, system status) are transmitted via Sub-GHz LoRa Mesh, while compressed thermal/visual keyframes ($320\times240$ JPEG @ $10\text{ fps}$) flow over the ESP-NOW / Wi-Fi mesh [RESEARCHED, INFERENCE]. The Surface Base Station receives the packets, parses JSON payloads, and updates the React.js Command Dashboard in real time [INFERENCE].

12. Technology Stack

13. AI/ML Justification

1. Thermal Human / Worker Detection Engine

• Exact Purpose: Identify trapped or unconscious mine workers in absolute darkness, dust, or smoke using body heat signatures [PS, RESEARCHED].
• Model Type: YOLOv8n fine-tuned on custom thermal infrared human datasets [RESEARCHED].
• Input: $320\times240$ single-channel LWIR thermal image feed from FLIR Lepton sensor [INFERENCE].
• Output: Bounding boxes $(x, y, w, h)$, class confidence score (worker), and spatial centroid coordinates [INFERENCE].
• Data Required: 1,500 annotated thermal image frames depicting humans sitting, lying down, partially occluded by debris, and standing in cold background environments [RESEARCHED, ASSUMPTION].
• Why AI is Necessary: Rule-based temperature thresholding fails because warm machinery motors, steam pipes, and ambient heat friction produce severe false positives [INFERENCE]. Deep learning extracts structural shape features in addition to heat intensity [INFERENCE].
• Evaluation Metrics: Mean Average Precision (mAP50 $> 0.85$), Recall ($> 0.90$), Inference Latency ($< 25\text{ms}$) [INFERENCE].
• Failure Mode & Mitigation: If thermal contrast is washed out due to mine underground fires ($>40^\circ\text{C}$ ambient), system falls back to IR night-vision camera feed with bounding box confidence flags [INFERENCE].
• Human Oversight: High bounding box detections ($>0.70$) automatically flag a visual alert on the surface dashboard; human rescue operators confirm before deploying personnel [INFERENCE].

2. Structural Hazard & Obstacle Classification Engine

• Exact Purpose: Identify gallery blockages, roof collapses, and standing water bodies in front of the rover [PS, INFERENCE].
• Model Type: Classical LiDAR point cloud height-field filtering + OpenCV geometric depth analysis [INFERENCE].
• Input: 2D/3D LiDAR distance arrays + visual depth map [INFERENCE].
• Output: Traversability costmap matrix ($0 = \text{clear}$, $255 = \text{impassable obstacle/wall}$) [INFERENCE].
• Why AI/Deep Learning IS NOT Used Here: Complex 3D point cloud segmentation deep neural networks require excessive memory and compute on edge devices [INFERENCE]. Geometric spatial thresholding (slope $> 20^\circ$, step height $> 10\text{ cm}$) is 100% deterministic, mathematically explainable, and significantly faster [INFERENCE].

14. Data Architecture & Offline/Low-Network Design

Local SQLite Database Schema

-- Sensor Telemetry Log Table
CREATE TABLE telemetry_log (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    timestamp DATETIME DEFAULT CURRENT_TIMESTAMP,
    ch4_concentration REAL NOT NULL, -- Methane %
    co_ppm REAL NOT NULL,            -- Carbon Monoxide PPM
    temp_celsius REAL NOT NULL,       -- Ambient Temperature
    humidity_pct REAL NOT NULL,      -- Relative Humidity
    battery_level REAL NOT NULL,     -- Rover battery %
    rssi_dbm INTEGER NOT NULL        -- Current active mesh link RSSI
);

-- Detected Trapped Miner Log Table
CREATE TABLE miner_detections (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    timestamp DATETIME DEFAULT CURRENT_TIMESTAMP,
    pose_x REAL NOT NULL,            -- Map relative X coordinate (meters)
    pose_y REAL NOT NULL,            -- Map relative Y coordinate (meters)
    confidence REAL NOT NULL,        -- YOLO confidence score (0.00 - 1.00)
    thermal_snapshot_path TEXT,      -- Local file path to saved JPEG
    status TEXT DEFAULT 'UNVERIFIED' -- UNVERIFIED / CONFIRMED / RESCUED
);

-- Mesh Relay Nodes Registry
CREATE TABLE mesh_nodes (
    node_id INTEGER PRIMARY KEY,
    dropped_timestamp DATETIME,
    battery_v REAL,
    mesh_hop_count INTEGER,
    status TEXT DEFAULT 'ACTIVE'     -- ACTIVE / OFFLINE / DEGRADED
);

Low-Network & Bandwidth Optimization Strategy

• Adaptive Multi-Tier Data Pipeline:
  1. Priority 1 (Critical Telemetry over LoRa 868MHz): Methane alerts, GPS/SLAM position, battery status, and panic triggers are packed into tiny 24-byte binary structures and broadcast every $1,000\text{ms}$. Transmission succeeds even at extremely low data rates ($1.2\text{ kbps}$) across long distances [RESEARCHED, INFERENCE].
  2. Priority 2 (Map & Image Data over ESP-NOW / Wi-Fi Mesh): Compressed 2D occupancy grid updates (delta compressed) and downsampled thermal keyframes ($320\times240$ JPEG @ $5\text{ FPS}$) are transmitted when Wi-Fi bandwidth is $>100\text{ kbps}$ [INFERENCE].
• Onboard Local Buffering & Re-Sync: If total wireless communication is completely lost (e.g., severe rock collapse blocks radio signal), the Jetson board buffers all LiDAR scans, high-res video, and gas logs to an onboard local SQLite database / ROSbag on NVMe storage [INFERENCE]. Once the rover backtracks into range of the nearest functional dropped node, the system executes an automated differential sync protocol [INFERENCE].

15. Security, Privacy & Governance

• Personal & Location Data Protection: Telemetry and thermal video feeds contain no personally identifiable information (PII) beyond anonymized human body heat signatures [INFERENCE]. All wireless transmissions over LoRa and Wi-Fi Mesh are encrypted using AES-128-GCM with pre-shared keys configured prior to deployment [RESEARCHED, INFERENCE].
• System Authentication & Access Control: Surface dashboard access requires role-based authentication (Admin / Rescue Operator / Inspector) via local JWT tokens to prevent unauthorized override of rover teleoperation controls [INFERENCE].
• Explosion Safety Compliance (DGMS / IS Standards):
  • Statutory Norm: Directorate General of Mines Safety (DGMS) mandates that all electrical equipment used in underground coal mines in India must comply with IS/IEC 60079-1 (Flameproof Enclosures "Ex-d") or IS/IEC 60079-11 (Intrinsic Safety "Ex-ia/Ex-ib") [RESEARCHED].
  • Hackathon Disclaimer: The student project built during the hackathon serves as a functional technology demonstration prototype [INFERENCE]. Real-world mine deployment requires formal certification from CIMFR (Central Institute of Mining and Fuel Research, Dhanbad) and DGMS approval [RESEARCHED, INFERENCE].
• System Liability & Human Oversight: The AI detection system acts purely as a decision-support tool [INFERENCE]. Final clearance for rescue team entry rests strictly with the Human Rescue Commander based on cross-validated sensor readings [INFERENCE].

16. MVP Scope (Hackathon Build)

+-----------------------------------------------------------------------------------+
|                            HACKATHON BUILD MVP SCOPE                              |
+-----------------------------------------------------------------------------------+
| MUST BUILD (Core Value Proposition)                                               |
|  [x] Custom tracked ground chassis with differential drive control.               |
|  [x] Multi-gas sensor array (MQ-4 Methane, MQ-7 CO, DHT22) with local alarm.      |
|  [x] FLIR Lepton 3.5 / AMG8833 thermal camera payload on Jetson Orin Nano.       |
|  [x] Edge AI YOLOv8 thermal human detection model running locally in real time.   |
|  [x] 2D LiDAR RPLiDAR A1M8 integration with ROS 2 slam_toolbox mapping.           |
|  [x] Automated servo-actuated droppable mesh repeater node mechanism (2 nodes).   |
|  [x] Web-based Surface Command Dashboard with live gauges, video, and 2D map.     |
+-----------------------------------------------------------------------------------+
| SHOULD BUILD (Differentiators - If Time Permits)                                  |
|  [/] ESP-NOW low-overhead image transmission protocol between nodes.              |
|  [/] Automated RSSI-triggered node deployment loop.                               |
|  [/] Audio-visual pan-tilt head for directional exploration.                      |
+-----------------------------------------------------------------------------------+
| NICE TO HAVE (Post-Hackathon Roadmap)                                             |
|  [ ] Micro-tethered aerial drone for high-roof gallery cavity inspection.         |
|  [ ] Full 3D point-cloud mesh reconstruction via GPU nvblox.                      |
|  [ ] CO2 and Optical Oxygen deficit sensors.                                      |
+-----------------------------------------------------------------------------------+
| DO NOT BUILD DURING HACKATHON (Out of Scope / Infeasible)                         |
|  [N/A] Official Flameproof / Intrinsic Safety (FLP/IS) certified steel housings. |
|  [N/A] Autonomous multi-rover swarm coordination engines.                         |
|  [N/A] Direct Through-The-Earth (TTE) sub-kilohertz voice transceivers.           |
+-----------------------------------------------------------------------------------+

17. Live Demo Simulation Plan

To deliver a compelling, authentic live demonstration to SIH judges without requiring access to an actual underground coal mine:

1. Simulated Mine Gallery Environment:
   • Construct a $10\text{-meter}$ dark, L-shaped mock tunnel using cardboard/plywood boxes, furniture, and cloth covers in the judging hall to simulate a GPS-denied underground cross-cut with a sharp $90^\circ$ turn [INFERENCE, ASSUMPTION].
2. Thermal Human Detection Simulation:
   • Place a team member or heating mannequin wrapped under blankets/boxes at the far end of the dark L-shaped tunnel [INFERENCE, ASSUMPTION].
   • Show live thermal feed on dashboard with YOLO bounding box pinpointing "TRAPPED MINER DETECTED" and plotting precise coordinates on the SLAM grid [INFERENCE].
3. RF Corner Obstacle & Automated Node Drop Demonstration:
   • Drive rover into the tunnel past the $90^\circ$ corner [INFERENCE].
   • Show surface dashboard RSSI meter dropping from $-50\text{ dBm}$ to $-82\text{ dBm}$.
   • Trigger the onboard servo mechanism live: the rover automatically drops a droppable mesh repeater node on the floor [INFERENCE].
   • Show dashboard RSSI immediately recovering to $-48\text{ dBm}$ and live video stream clearing up instantly [INFERENCE].
4. Gas Leak & Alarm Simulation:
   • Expose the $MQ-4$ Methane sensor to a controlled, safe butane lighter gas vapor [INFERENCE].
   • Show real-time dashboard gas gauge spiking into red zone ($>1.5%$), triggering the surface audible siren and updating safety status to "DANGER: HIGH METHANE" [INFERENCE].

18. Implementation Roadmap

Phased Roadmap (Target Deadline: 30 September 2026) [PS]

Phase 1: Hardware Design & Sensor Integration (Weeks 1 - 3)
├── Select & procure tracked chassis, motor drivers, ESP32, gas sensors, LiDAR.
├── Design & 3D print servo-actuated droppable node magazine.
└── Validate micro-ROS UART communication between ESP32 and Jetson.

Phase 2: SLAM Mapping & Edge AI Model Training (Weeks 4 - 6)
├── Collect & annotate thermal human image dataset (FLIR/AMG8833).
├── Train & export YOLOv8n model to TensorRT FP16 format on Jetson.
└── Configure ROS 2 slam_toolbox and RPLiDAR drivers for 2D occupancy mapping.

Phase 3: Wireless Mesh Development & Breadcrumb Logic (Weeks 7 - 9)
├── Develop ESP-NOW / LoRa mesh protocol for multi-hop packet relaying.
├── Implement automated RSSI monitoring & servo drop trigger script.
└── Conduct line-of-sight and corner RF attenuation benchmark tests.

Phase 4: Surface Command Dashboard & System Fusion (Weeks 10 - 11)
├── Build React.js dashboard with WebSockets streaming, Chart.js, and Leaflet.
├── Integrate MQTT telemetry parsing & SQLite local logging.
└── Perform end-to-end dry runs in dark, obstructed corridors.

Phase 5: Field Testing, Refinement & SIH Presentation Prep (Weeks 12 - Final Deadline: Sep 30, 2026)
├── Conduct stress tests under extreme smoke, dust, and gas leak simulations.
├── Prepare live demo kit, fallback offline video logs, and technical dossier.
└── Final Code Freeze and SIH Demo Polish.

19. Testing & Validation Plan

• Functional Sensor Testing: Expose $MQ-4$ and $MQ-7$ sensors to known safe gas concentrations in a controlled chamber; verify response time ($T_{90} < 15\text{s}$) and accuracy against reference digital meters [INFERENCE].
• SLAM Mapping Accuracy Test: Drive rover through a measured $20\text{-meter}$ corridor with obstacle blocks; compare ROS 2 generated grid map dimensional error against laser measure tape ($< 5\text{ cm}$ target variance) [INFERENCE].
• AI Model False Positive / False Negative Validation:
  • Dataset: Evaluate YOLOv8n thermal model on 200 unseen thermal frames containing hot motors, light bulbs, steam, and human bodies [INFERENCE].
  • Pass Criteria: Precision $>88%$, False Positive Rate $<5%$, False Negative Rate $<3%$ on human body class [INFERENCE].
• Wireless Mesh Relay Drop Test: Drive rover past two consecutive $90^\circ$ concrete walls; verify that node drop triggers automatically when RSSI $< -75\text{ dBm}$ and video feed latency remains $< 300\text{ms}$ over 3 hops [INFERENCE].
• Power & Thermal Endurance Test: Run rover continuously under full sensor + LiDAR + Jetson compute load for 2 hours; verify motor driver temperature remains $< 80^\circ\text{C}$ and battery voltage stays above cutoff threshold [INFERENCE].

20. Risk Register

21. Cost & Scalability

Hackathon Prototype Cost (Single System)

• Tracked Robot Chassis + Planetary Gear Motors: $\text{₹}12,000$ [ASSUMPTION]
• NVIDIA Jetson Orin Nano Developer Kit (8GB): $\text{₹}24,000$ [RESEARCHED]
• RPLiDAR A1M8 2D LiDAR: $\text{₹}8,500$ [RESEARCHED]
• Thermal Sensor (FLIR Lepton 3.5 / AMG8833) + IR Night Cam: $\text{₹}14,000$ [RESEARCHED]
• ESP32 Microcontrollers + Gas Sensors ($MQ-4, MQ-7, DHT22$): $\text{₹}3,500$ [RESEARCHED]
• Droppable Mesh Nodes x 3 (ESP32 + Li-Ion + 3D Casing + Servos): $\text{₹}4,000$ [INFERENCE]
• Battery ($12\text{V }10\text{Ah }LiFePO_4$) & Power Regulators: $\text{₹}5,500$ [INFERENCE]
• Total Prototype Cost: $\sim\mathbf{\text{₹}71,500}$ [INFERENCE]

Production System Cost (Flameproof / Industrial Scale)

• DGMS Certified FLP Cast-Steel Chassis & Explosion-Proof Motors: $\text{₹}3,500,000$ [INFERENCE]
• Industrial Grade Ouster 32-Channel 3D LiDAR + FLIR Boson Core: $\text{₹}1,200,000$ [INFERENCE]
• ATEX/DGMS Intrinsic Safe Sensor Suite & Sub-GHz Mesh Nodes: $\text{₹}400,000$ [INFERENCE]
• Total Production Unit Cost: $\sim\mathbf{\text{₹}5,100,000\text{ per unit}}$ [INFERENCE]
• Scalability Vision: Production costs drop by $40%$ under centralized procurement by Coal India Ltd (CIL) for deployment across 400+ underground coal mines nationwide [INFERENCE].

22. Differentiation

23. Future Scope

1. Integration of Tethered Micro-Drone: Add an automated pop-up micro-quadcopter powered via a light thin tether from the rover to fly into elevated roof collapse cavities and vertical ventilation shafts [INFERENCE].
2. GPU-Accelerated 3D Mesh Reconstruction (NVIDIA nvblox): Upgrade 2D mapping to full volumetric 3D TSDF signed distance field meshes for VR-based surface rescue planning [RESEARCHED].
3. Bio-Sensing Gas Sensors: Integrate miniaturized photoionization detectors (PID) for ultra-low PPM detection of volatile organic compounds and toxic benzene gases [INFERENCE].

24. Assumptions & Open Items Log

Consolidated Assumptions

• [ASSUMPTION-1]: The student team has access to an NVIDIA Jetson Orin Nano or Jetson Nano for edge AI execution.
• [ASSUMPTION-2]: A 2D LiDAR (RPLiDAR A1M8) provides sufficient spatial geometry for 2D SLAM in rectangular mine galleries.
• [ASSUMPTION-3]: Low-resolution thermal sensors (FLIR Lepton / AMG8833) combined with transfer-learned YOLO offer acceptable human body detection at $1\text{--}5\text{ meter}$ ranges.
• [ASSUMPTION-4]: Physical live demo can be convincingly executed using a dark indoor L-shaped mock tunnel setup.

Consolidated Items Needing Validation

• [NEEDS VALIDATION-1]: Sensor calibration drift rate of $MQ-4$ and $MQ-7$ sensors in actual high-humidity ($>90%$) underground coal mine atmospheres.
• [NEEDS VALIDATION-2]: Maximum transmission distance of 2.4GHz ESP-NOW mesh frames around damp, jagged coal seam corners before signal degradation requires a node drop.
• [NEEDS VALIDATION-3]: Formal approval process timeline and testing cost for CIMFR/DGMS flameproof certification of student robotics hardware.

Final Recommendation

BUILD THE PROPOSED HARDWARE SYSTEM. The hybrid tracked ground rover featuring an automated droppable mesh node mechanism, thermal YOLO detection, ROS 2 LiDAR SLAM, and WebSockets command dashboard is technically sound, perfectly aligned with SIH requirements, highly feasible for a 4–6 student team, and addresses the critical gap of subterranean signal loss [INFERENCE].

Recommended Stack

• Hardware: Tracked Chassis + Planetary Motors + NVIDIA Jetson Orin Nano + ESP32 + RPLiDAR A1M8 + FLIR Lepton 3.5 [RESEARCHED, ASSUMPTION].
• Software: Ubuntu 22.04 LTS + ROS 2 Humble + slam_toolbox + TensorRT YOLOv8n + React.js + Node.js [RESEARCHED, INFERENCE].

Biggest Dependency

Availability and configuration of the NVIDIA Jetson edge AI board and dual-camera thermal sensor payload [INFERENCE].

Biggest Risk

RF signal attenuation underground causing complete video frame loss if the automated breadcrumb node drop fails to deploy [RESEARCHED, INFERENCE].

Biggest Technical Challenge

Achieving real-time ROS 2 SLAM mapping, thermal YOLO inference, and multi-hop wireless packet relaying simultaneously on a single embedded compute budget without overheating [INFERENCE].

First 5 Actions

1. Procure tracked chassis, high-torque gear motors, ESP32 microcontrollers, and RPLiDAR A1M8 [INFERENCE].
2. Flash Ubuntu 22.04 LTS and ROS 2 Humble onto NVIDIA Jetson Orin Nano [RESEARCHED].
3. Build and test 2D SLAM mapping using slam_toolbox on Jetson with RPLiDAR [RESEARCHED].
4. Design and 3D print the servo-driven droppable mesh node magazine and write ESP-NOW mesh firmware [RESEARCHED, INFERENCE].
5. Train YOLOv8n on thermal human image dataset, convert to TensorRT, and integrate with live dashboard feed [RESEARCHED, INFERENCE].

Validation Before Coding

Verify that the ESP32 microcontrollers can reliably exchange ESP-NOW peer-to-peer data packets across 3 hops while reading analog gas sensor values on hardware timer interrupts [INFERENCE].

Buildability Score /100

88/100 (High buildability for an energetic 4–6 student engineering team within a hackathon timeframe) [INFERENCE].

Confidence

92% Confidence in proposed technical architecture and problem solution alignment [INFERENCE].

Research Sources