Minefield Swarm – Information Flow Architecture

Detection-agnostic multi-drone system for underground minefield escort and mapping.

architecture · information flow · swarm robotics
System Overview Concept

This document describes how information flows inside a four-drone minefield escort system:

  • 3× Scanning Drones — low-altitude platforms that sense underground mines and poles, do light local processing and send map tiles + health.
  • 1× Master Drone — swarm “brain” that fuses maps, plans coverage, computes a safe corridor, manages formation and human interaction.
  • User (Human Walker) — walks strictly inside the safe corridor, issues high-level commands only (start, stop, abort, slow, etc.).
  • Environment — underground mines, poles, and ground material that influence sensor readings.

The architecture is deliberately detection-agnostic: any combination of mine-sensing hardware (ultrasonic, inductive, capacitive, etc.) can be plugged in as long as each scanning drone outputs location-stamped evidence and small local map patches.

Underground mines
Swarm robotics
Occupancy mapping
Safe path planning
Human–swarm interaction
Mission Modes State Machine

The master drone runs a simple but explicit mission state machine:

  1. BOOT / SELF-CHECK — verify sensors, comms, battery, localization sanity.
  2. FIELD-SCAN MODE — human waits at start, swarm explores and builds map.
  3. PATH-READY / WAIT — uncertainty below threshold; safe corridor exists but user has not started walking.
  4. ESCORT MODE — user walks; drones maintain and update safe corridor around them.
  5. ABORT / FAIL-SAFE — any hard safety violation or serious health issue.

All commands from the user are interpreted as requests to transition between these modes; safety rules always win over user intent.

High-Level Information Flow
Block Diagram A — System Level Visual
Environment → Scanning Drones → Master Drone → User
High-level information flow Environment Underground mines Poles & obstacles Ground material 3× Scanning Drones Sensors + local processing Outputs to Master Local map tiles (mine/free/unknown) Evidence events (pose, conf, type) Health summary (battery, sensors) Master Drone • Global map fusion • Coverage / scan planner • Safe path / corridor planner • Formation & swarm controller • HRI & mission state machine User (Human Walker) Walks in safe corridor Gestures / voice commands Receives state & warnings Physical influence Local tiles + health Scan / escort waypoints Safe corridor, guidance High-level commands
Environment
Scanning drones
Master drone
User

Diagram A: High-level information and command flow. The master drone is the only brain; scanning drones and the human never bypass its safety logic.

Node Details

Environment / Ground

The environment block is not “data” that drones receive; it is the physical world:

  • Underground mines at unknown coordinates.
  • Poles and other obstacles.
  • Ground material which affects sensor penetration and noise.

Sensors on each drone interact with this environment to generate raw signals.

Scanning Drones (×3)

Each scanning drone carries navigation and mine-sensing hardware and runs a lightweight processing stack:

  • IMU, barometer, GPS/UWB/optical flow for pose estimation.
  • Mine-sensing payload — inductive, capacitive, ultrasonic, IR, etc.
  • Filtering and calibration of sensor signals.
  • Detection logic that turns raw data into evidence events with pose, sensor type, and confidence.
  • Integration of evidence into a local occupancy / evidential map tile.
  • Health monitoring (battery, sensor OK flags, link quality).

The scanning drones never run global planning; they simply generate high-quality local information and report health.

Master Drone

The master drone is the swarm brain. It performs heavy computation:

  • Global Map Fusion — fuses prior layout + all local tiles into a global belief map and an uncertainty map.
  • Coverage / Scan Planner — decides which regions still need sensing and assigns scan strips to scanning drones.
  • Safe Path & Corridor Planner — computes a safe route and a corridor with ≥ 1 m clearance from any mine cluster.
  • Formation / Swarm Controller — converts the corridor into per-drone trajectories (beacon, flankers, overwatch).
  • HRI & Mission State Machine — interprets user commands, enforces safety, and switches between mission modes.

User (Human Walker)

The user interacts through simple, robust channels:

  • Gesture or voice commands mapped to a small command set: start, stop, pause, resume, slow, abort.
  • Visual/audio feedback: LEDs, beeps, and characteristic drone motion to encode state and warnings.
  • Physical rule: user stays inside the corridor provided by the drones.

The user never sends direct flight commands; they only request mode changes. Safety always has higher priority than user intent.

Scanning Drone Internal Pipeline
Block Diagram B — Per-Drone Processing Visual
From environment → evidence events → local map tiles + health.
Scanning drone internal pipeline Environment Mines, poles, ground Navigation Sensors IMU, baro, GPS/UWB, OF Mine-Sensing Module inductive / capacitive / etc. Pre-processing Filtering, calibration, time sync Detection Logic thresholds / small ML / patterns Evidence Events (x,y,pose, sensor, conf, t) Local Map Tile small occupancy / evidential grid Health Summary battery, sensors, links

Diagram B: Each scanning drone converts raw sensor data into evidence events, then into a compact local map tile and a health summary streamed to the master drone.

Master Drone Internal Architecture
Block Diagram C — Master Drone Visual
All heavy fusion, planning and safety logic lives here.
Master drone internal architecture Local Map Tiles + Health from all scanning drones Prior Layout Map converted to prior probability grid User Commands gestures / voice / kill switch Global Map Fusion fuse prior + local tiles output occupancy + uncertainty Coverage / Scan Planner allocate regions to scanners Safe Path & Corridor A*/D* on safe cells + inflation Formation Controller assign roles + trajectories HRI & Mission State Machine interpret commands, enforce safety State & Warnings to User LEDs, beeps, motion patterns scan waypoints to scanners escort trajectories to all drones
Mapping & planning logic
HRI / safety logic

Diagram C: Inside the master drone. All global reasoning and safety decisions are centralized here, while scanning drones remain lightweight and interchangeable.

Information Flows (Text Specification)

For implementation or simulation, the info-flow can be summarised as directed edges between logical nodes:

  • Environment → ScanningDrone[i].Sensors: physical interaction of mines/poles with sensor hardware.
  • ScanningDrone[i].Sensors → PreProcessing → Detection → EvidenceEvents: conversion from raw signals to discrete evidence.
  • EvidenceEvents → LocalMapTile: integration into local occupancy / evidential grid.
  • LocalMapTile + Health → Master.GlobalMapFusion: streaming of compressed tiles and health summaries.
  • PriorLayout → Master.GlobalMapFusion: prior probability map from the given A4 layout.
  • GlobalMapFusion → {CoveragePlanner, PathPlanner}: distribution of occupancy and uncertainty maps.
  • CoveragePlanner → ScanningDrone[i]: scan waypoints and role assignments during FIELD-SCAN.
  • PathPlanner → FormationController: centerline and corridor width for the human.
  • FormationController → AllDrones: per-drone trajectories and separation constraints during ESCORT.
  • UserCommands → HRI → MissionStateMachine: high-level intents filtered and checked for safety.
  • MissionStateMachine → AllModules: mode changes (BOOT, SCAN, PATH-READY, ESCORT, ABORT).
  • Master.State + LocalMapSlice → User: state indicators and local map information encoded as LEDs, sounds, and motion.

This HTML page can act as both project documentation and a living spec while you iterate on sensor choices and detection algorithms.