AUTONOMOUS SEARCH & RESCUE · OFFLINE FIRST

AERIAL DISASTER ASSISTANCE / 01

When the network
goes dark, the mission
keeps moving.

SAADAXA is an autonomous aerial disaster assistance and location-tracking system engineered to find survivors, preserve geographic context and relay critical coordinates when satellite navigation and public communications are compromised.

GNSS-denied navigationOn-device edge AILong-range fallback
✳SAADAXA / UAV 102ILLUSTRATIVE MISSION RENDER
MISSION SYSTEMEDGE AUTONOMY
POSITIONING MODEANCHOR + SLAMLINK PRIORITYCOORDINATE RELAY
01 — 20 FIGURES / TECHNICAL WALKTHROUGHDESIGNED FOR INFRASTRUCTURE-COMPROMISED ZONES ↘
680mmHexacopter wheelbase
6×BLDC propulsion units
865MHzNarrowband telemetry
22–26minSpecified flight endurance
3 nodesDrone · Server · App

01 THE MISSION PROBLEM

Designed for the moment
infrastructure disappears.

Disaster zones are difficult environments for connected robotics. SAADAXA brings aerial sensing, local processing, geographic recovery and field coordination into one offline-capable system.

01 / FRAGMENTED RESPONSE
⌘

Disconnected tools

Conventional UAV deployments can operate as isolated endpoints. SAADAXA links the aircraft, a dedicated local server and the field operator terminal into one coordinated workflow.

RESPONSE → UNIFIED DSA ARCHITECTURE
02 / NETWORK OUTAGE
⌁

No public network required

Cellular towers, internet backhaul and power infrastructure may be unavailable. The design keeps critical telemetry and survivor coordinates on local and direct radio paths.

RESPONSE → OFFLINE MULTI-TIER LINKS
03 / SATELLITE DENIAL
◎

Coordinates must survive GPS loss

Visual odometry alone provides relative movement. SAADAXA anchors that movement to the last valid geographic fix and projects it back into WGS84 coordinates.

RESPONSE → ENU ANCHOR + RE-ANCHOR
04 / BANDWIDTH COLLAPSE
◌

Send the data that matters

Video is not suitable for narrowband LoRa. When Wi-Fi degrades, video and audio are suspended so small, high-priority position and telemetry packets can continue.

RESPONSE → CHANNEL ARBITRATION
“

The platform is not designed around uninterrupted connectivity. It is designed around mission continuity when connectivity is interrupted.

SAADAXA · SYSTEM DESIGN INTENT
12°58′ N
77°39′ E

02 SYSTEM ARCHITECTURE

One mission.
Three connected nodes.

The triangular Drone–Server–App architecture separates airborne autonomy, field-side coordination and operator control—while keeping the complete workflow usable without public internet infrastructure.

Patent Figure 1: Drone–Server–App network topology
FIG. 01 DSA NETWORK TOPOLOGY

AERIAL EDGE PLATFORM

The aircraft sees, computes and flies.

The hexacopter carries the sensor suite, onboard Raspberry Pi 5 compute, Pixhawk flight control, GNSS/visual navigation and dual-channel communications.

  • Pixhawk 2.4.8 / ArduCopter 4.x
  • Raspberry Pi 5 / 8 GB RAM
  • RealSense D435i + thermal sensor
  • Wi-Fi + 865 MHz LoRa / Vivaldi
Sense→Process→Navigate
COMMUNICATION CONTRACTCONTROL + TELEMETRY ≠ VIDEOMAVLink / WebSocket / RTSP / LoRa payloads

03 AIRBORNE PLATFORM

A six-rotor frame built
around the sensor mission.

The specification describes a symmetrical carbon-fibre hexacopter, with a central avionics hub and six radial arms. The lower equipment shelf is vibration-isolated to protect optical navigation instruments.

Patent Figure 13: isometric and top plan views of the hexacopter frameFIGURE 13 / AIRFRAME DESIGN
PLATFORM SPECIFICATIONUAV / 102
Frame configurationCarbon-fibre hexacopter
Motor-to-motor wheelbase680 mm
Propulsion6 × 650 KV BLDC motors
Electronic speed controllers30 A BLHeli32
Motor command protocolDSHOT300
Flight controllerPixhawk 2.4.8
Firmware / estimatorArduCopter 4.x / EKF3
Power pack6S · 10,000 mAh LiPo
✳

Six-rotor geometry supports motor-mix compensation in a single-motor anomaly scenario, followed by an automated return-to-launch profile.

01Central hubDual deck plates, internal standoffs and the avionics stack.
02Radial armsSix carbon-fibre booms with shielded motor wiring routes.
03Isolated shelfViscoelastic damping rings reduce rotor vibration reaching optical sensors.
04Retractable gearGround stability with clearance for downward optical and thermal sensors.

04 COMPUTE + SENSOR FUSION

Two control layers.
One fused picture.

The Raspberry Pi 5 handles high-level autonomy, visual processing and edge intelligence. The Pixhawk handles real-time flight stabilization, state estimation and motor mixing.

HIGH-LEVEL AUTONOMY
RP5

Raspberry Pi 5

Quad-core 64-bit ARM Cortex-A76 · 2.4 GHz · 8 GB RAM

FastAPI servicesYOLOv8-nanoORB-SLAM3ENU projectionLink manager
Supervisory compute / edge AI / coordinate recovery
UART MAVLink 2.0 VISION_POSITION_ESTIMATE
⇄
REAL-TIME FLIGHT CONTROL
PX2.4.8

Pixhawk 2.4.8

ArduCopter 4.x · EKF3 · 32-bit flight controller

IMUAttitude loopMotor mixingFailsafesESC outputs
Stabilization / navigation estimate / propulsion control
MULTI-RATE ESTIMATION

Sensor update schedule

Frequencies as described in the specification
INPUTROLE IN THE SYSTEMRATECONSUMER
IMUAttitude and acceleration propagation400 HzEKF3 / Pixhawk
BarometerVertical altitude reference50 HzEKF3 / Pixhawk
Hereflow optical flowLow-altitude translational stability50 HzEKF3 / Pixhawk
RGB-D / visual odometryDepth, visual features and metric displacement30 Hz outputRaspberry Pi / EKF3
Proximity / IRObstacle range and clearance20 HzFlight / landing logic
RTK GNSSAbsolute geographic fix5–10 HzEKF3 / anchor manager
Thermal imagingLocalized heat-signature verificationFrame-synchronisedEdge detection pipeline
Patent Figure 2: sensor and peripheral architecture
FIG. 02 — UAV SENSOR LAYERS
Patent Figure 3: EKF3 sensor fusion input matrix
FIG. 03 — EKF3 SENSOR FUSION

06 ONBOARD EDGE INTELLIGENCE

Find the person.
Calculate the position.

Survivor detection is processed on the aircraft. Visual structure proposes a human candidate, thermal data corroborates the signature, and depth converts the detection into a spatial offset.

Illustrative hexacopter using depth and thermal sensing above earthquake rubble
THERMAL CANDIDATE
ONBOARD INFERENCE STACK

YOLOv8-nano
NCNN / INT8

Optimised CPU-based inference on the Raspberry Pi 5. The specification describes a compact model designed to avoid cloud dependency and dedicated GPU acceleration.

3.2 MB
Quantised model footprint stated in the specification
~15 FPS640 × 640 inference input
<300 msRaw frame to WebSocket push
DETECTION → GEOLOCATION → NOTIFICATIONTARGET PIPELINE / 09 STAGES
01Frame captureSynchronised RGB + depth acquisition
02YOLO inferenceHuman candidate bounding boxes
03Confidence filterReject weak detections
04Thermal matchRegister and corroborate heat signature
05Depth projectionExtract distance and 3D offset vector
06World transformUAV frame → ENU → WGS84
07De-duplicateMerge repeat spatial detections
08WebSocket pushSend priority target event to server
09Operator alertPlot target on the tactical map
VISUAL + DEPTH + THERMAL→3D TARGET OFFSET→ABSOLUTE COORDINATE→FIELD ALERT
ASYNC INFERENCEThreadPoolExecutor inside an asyncio event loop keeps model inference from blocking the main FastAPI service loop.
HARDWARE VIDEO ENCODINGv4l2h264enc is described for H.264 acceleration, with encoding latency reduced by approximately 40% versus software encoding.
CPU OPERATING MODEThe CPU governor is specified to lock to performance mode at system boot.
SIMULATION RESULTS STATED IN THE PATENT SPECIFICATION

Reported detection metrics by altitude

These are specification-reported simulation figures, not independent test results from this website.

30 m AGLPrecision / Recall / mAP@0.5
84.381.283.1
50 m AGLPrecision / Recall / mAP@0.5
70.468.169.2
PRECISIONRECALLmAP@0.5
Patent Figure 6: nine-stage edge AI processing and alert pipeline
FIG. 06 — EDGE-AI DETECTION PIPELINE
Patent Figure 11: search strip flight pattern with mid-strip detection event handling
FIG. 11 — SURVEY PATH / EVENT HANDLING

07 RESILIENT COMMUNICATIONS

Two radios.
Two different jobs.

The design deliberately separates high-bandwidth media from low-bandwidth critical coordinates. It does not try to push video through LoRa; it changes the data plan when the link changes.

LIVE LINK-ARBITRATION DEMO

Choose a link condition

CHANNEL A / PRIMARY⌁

Wi-Fi wideband

High-throughput path for real-time media and full command interaction when signal quality is healthy.

  • 2.4 / 5.8 GHz link
  • H.264 video via GStreamer RTSP
  • Opus two-way audio at 32 kbps
  • Telemetry and flight commands
CHANNEL B / FALLBACK⌖

LoRa + Vivaldi

Low-rate, long-range channel dedicated to the smallest mission-critical payloads—not video or voice.

  • RYLR998 / SX1276 family · 865 MHz
  • SF9 · BW 125 kHz · CR 4/5
  • Compressed payloads under 240 bytes
  • Multi-hop aerial mesh forwarding
PASSIVE RF DIRECTIVITY

Shape the beam.
Do not spend more power.

The end-fire planar tapered-slot Vivaldi antenna focuses radio-frequency energy toward the ground command node. Its passive directive gain is specified at +4 to +7 dBi, without additional electrical power draw.

+4 to +7 dBiSpecified directive gain2–5 kmStated damaged-environment range

For comparison, the specification states approximately 1–3 km for a stock whip in damaged or obstructed conditions (8–10 km theoretical unobstructed), versus 2–5 km with the directional Vivaldi in damaged environments.

Patent Figure 9: omnidirectional stock whip versus end-fire Vivaldi antenna radiation patternFIG. 09 / RADIATION PATTERN COMPARISON
!
Bandwidth rule

When the Wi-Fi RSSI falls below −75 dBm, video and audio are suspended. The 865 MHz link carries compressed coordinates and core telemetry only. If both channels fail to receive ground-station heartbeat acknowledgements for 60 seconds, the system initiates return-to-launch.

Patent Figure 7: Wi-Fi and LoRa failover state machine
FIG. 07 — DUAL-CHANNEL FAILOVER
Patent Figure 8: LoRa radio and Vivaldi antenna layers
FIG. 08 — LONG-RANGE RADIO LAYERS
Patent Figure 10: channel bandwidth allocation
FIG. 10 — BANDWIDTH ALLOCATION

08 MISSION EXECUTION

Survey systematically.
Respond intelligently.

The mission planner decomposes an operator-defined search area into parallel strips. A mid-strip detection interrupts the pattern for confirmation, then the planned survey can resume.

SEARCH ZONE / STRIP PATTERNEVENTRESUME PATH →
20–30% OVERLAPALTITUDE + HFOV BASED SPACING
01
Define the survey boundary

Operator sets the target area and mission limits.

02
Calculate strip width

Altitude and camera horizontal field of view determine ground swath width.

03
Apply overlap

Adjacent legs are spaced to maintain a stated 20–30% overlap margin.

04
Execute parallel strips

A snake-like raster path covers the defined search zone.

05
Pause for a survivor event

The UAV stabilises over the detected target for a confirmation hover, then resumes the survey path.

Patent Figure 11: strip decomposition and event management
FIG. 11 — STRIP-DECOMPOSITION SURVEY
Patent Figure 16: fail-safe trigger and response matrix
FIG. 16 — CONTINGENCY RESPONSE MATRIX

09 ACTIVE RESCUE INTERACTION

From locating a survivor
to reaching them.

The system extends beyond passive observation. The proposed platform includes two-way acoustic relay and a centrally mounted modular payload release mechanism.

◖
VOICE / OPUS
TWO-WAY ACOUSTIC RELAY

Rescue teams can speak.
Survivors can respond.

Audio is captured at 16 kHz PCM through ALSA / PyAudio, encoded with Opus at 32 kbps and relayed over WebSocket frames. The reverse path delivers operator audio to the onboard speaker.

<400 msWi-Fi voice latency statedWi-Fi onlyDisabled on narrowband LoRa
Patent Figure 14: underside servo-actuated payload release assemblySERVO / QUICK RELEASE
UNDERSLUNG PAYLOAD MECHANISM

Deliver small essentials
at the target location.

A servo actuator operates a quick-release locking latch under the central hub. The modular mount is described for sensor pods, medical-aid containers or emergency beacon transponders.

Central axisUnderslung mountingCommandedAutonomous or operator input

10 AUTONOMOUS TOUCHDOWN

Do not land on a guess.
Verify the surface.

Depth point clouds are used to identify a planar landing patch, check slope and obstacles, and verify low lateral drift before the final descent.

∠
<15°Maximum landing-surface inclination
◇
0.3 mObstacle height threshold
◎
1.0 mRadial touchdown clearance envelope
↔
±5 mExpanded alternate landing search radius
↓
<0.5 m/sHorizontal drift limit for final descent
01Capture depth point cloudDownward RealSense D435i data
↓
02Fit candidate landing planeEvaluate surface angle and obstacles
↓
Surface clear?Slope under limit + obstacle envelope clear
NOHold altitude · scan alternate zone
YESVerify drift + clearance
↓
03Authorise final 3 m descentTouchdown only after clearance confirmation
Patent Figure 12: landing zone planar verification flowchartFIG. 12 / AUTONOMOUS LANDING CLEARANCE

11 POWER + ENGINEERING ENVELOPE

Designed around a
real flight budget.

The patent specification describes a 6S, 10 Ah lithium-polymer battery, a nominal 348 W hover-search power estimate and 22–26 minutes of flight endurance.

NOMINAL HOVER / SEARCH
348 W

Estimated total platform power consumption stated for a standard hover-search profile.

Propulsion / 6 motors Avionics / compute + sensors Communications
STORED ENERGY
6S / 10 Ah

Lithium-polymer battery pack with nominal 22.2 V terminal voltage.

6-CELL PACK
22–26 MINUTES / SPECIFIED ENDURANCE
COST ESTIMATE IN SPECIFICATION
₹1.2–1.5 L

Approximate build-cost range mentioned in the supplied patent document.

Hot-swap battery bayField-charging accessoriesIntegrated avionics / payload
ENGINEERING NOTE

The document describes hot-swap battery architecture and field charging as mitigations for the stated endurance constraint. Actual flight time will vary with payload, wind, propulsion setup, battery condition and mission profile.

12 FAIL-SAFE DESIGN

Every critical fault
has a response path.

Hardware-level ArduCopter failsafes and software supervisors govern degraded navigation, link loss, low battery, motor anomalies and geofence events.

TRIGGER CONDITIONAUTONOMOUS RESPONSEPOLICY / THRESHOLD
GNSS tracking degradedSwitch to anchored visual-inertial navigationHDOP > 5.0 / loss of 3D fix
Visual tracking lostEnter stabilised position-hold stateSLAM / EKF3 health alert to app
GNSS reacquiredBlend pose, reset drift, establish new anchorHDOP < 2.0
Wi-Fi RSSI degradesMove critical telemetry to LoRaBelow −75 dBm
Both channels unacknowledgedAutonomous return-to-launch60-second heartbeat timeout
RC link lostAutonomous return-to-launchFS_THR_ENABLE = 1 · over 2 s
GCS heartbeat absentReturn-to-launch failsafeFS_GCS_ENABLE = 1 · over 5 s
Battery low warningPush notification and initiate return profileFigure 16 indicates below 20%
Battery criticalImmediate controlled landingBATT_FS_LOW_ACT = 1 · below 15% in detailed text
Motor RPM anomalyMotor-mix compensation, then RTLSingle-motor failure indication
Geofence breachPosition hold; RTL if unresolved10-second correction window
The 5-second ground-station heartbeat failsafe and the 60-second combined-link watchdog are separate conditions described in different parts of the specification.
Patent Figure 15: power budget and battery capacity diagram
FIG. 15 — POWER BUDGET
Patent Figure 16: failsafe trigger and response matrix
FIG. 16 — FAILSAFE MATRIX

13 SOFTWARE + FIELD INTERFACE

Local processing.
Shared situational awareness.

A dedicated local server bridges flight messages, aggregates detections and distributes the current mission picture to the Android field application.

01AIRCRAFT
Raspberry Pi 5ArduCopter / PixhawkLCM channel managerSensor + AI pipeline
MAVLink 2.0 / UART↓Wi-Fi + LoRa paths
02LOCAL SERVER
FastAPI async backendMAVLink bridgeWebSocket hub · 2 HzGStreamer / H.264 RTSPDatabase event logSpatial de-duplication
WebSocket state stream↓RTSP video / audio frames
03FIELD TERMINAL
Android / KotlinGoogle Maps SDKExoPlayerTelemetry dashboardCommand dispatchVoice relay controls
Patent Figure 17: inter-module circuit and communication diagram
FIG. 17 — INTER-MODULE CONNECTIONS
Patent Figure 18: detailed hardware wiring and avionics schematic
FIG. 18 — HARDWARE / PIN-LEVEL SCHEMATIC
USB 3.0RealSense RGB-D stream to the onboard processor
UART / MAVLink 2.0Processor ↔ Pixhawk telemetry and setpoints
I²C / SPI / UARTFlight controller peripheral navigation sensors
DSHOT300Digital throttle control to the ESCs

14 CORE METHOD

From signal loss
to a geolocated alert.

The independent method claim describes the full chain: monitor satellite integrity, latch an anchor, estimate local displacement, recover absolute coordinates, detect a subject and transmit the target location.

Redrawn Figure 20: method for GNSS-denied navigation and target localization
FIG. 20 / REDRAWN FROM SPECIFICATION TEXT
INPUTGNSS + visual + depth + thermal
TRANSFORMAnchor → ENU → WGS84
OUTPUTSurvivor geolocation → radio relay

15 OPERATIONAL USE CASES

Built for environments
that break the usual playbook.

The specification identifies disaster response, security, remote monitoring, humanitarian operations and infrastructure inspection as potential industrial applications.

01 / DISASTER MANAGEMENT

Earthquake, flood
and landslide response

Aerial mapping, survivor coordinate relay and field coordination where infrastructure is damaged.

NDMA · NDRF · SDRF
02 / DEFENCE + BORDER

Surveillance in
GNSS-degraded zones

Autonomous reconnaissance and direct mesh telemetry without dependence on commercial cellular networks.

DRDO · ARMY · BSF
03 / FOREST + WILDLIFE

Thermal perimeter
and heat-signature scans

Remote forestry survey and detection of heat signatures in areas with limited communications.

FOREST DEPARTMENTS
04 / URBAN SEARCH + RESCUE

Navigate debris
and collapsed structures

Depth-aware inspection, survivor localization and tactical communication across rubble fields.

MUNICIPAL FIRE SERVICES
05 / AD-HOC COMMUNICATIONS

Bridge isolated
field teams

Use aerial nodes as multi-hop relay points when terrestrial networks are unavailable.

OFF-GRID FIELD OPERATIONS
06 / CRITICAL INFRASTRUCTURE

Inspect remote
or damaged assets

Potential applications include bridges, dams, pipelines and high-voltage transmission infrastructure.

INSPECTION + ASSESSMENT

16 PATENT SCOPE

The invention, captured
in ten claims.

The supplied specification is titled “Autonomous Aerial Disaster Assistance and Location Tracking System and Its Method for the Same.” The claims cover an integrated UAV system, its supporting functions and the satellite-denied navigation method.

01Independent system claimUAV + anchor-based visual-inertial geolocation + multi-channel telemetry arbitration

A multi-rotor platform, propulsion and flight control, central computing, satellite-reliant / anchored visual-inertial mode switching, ENU coordinate recovery, visual-depth + thermal survivor detection, and wideband/narrowband communications with directional antenna fallback.

02Automatic re-anchoringRestore a valid satellite fix and reset drift

Blend visual-inertial displacement with the re-established satellite fix, reset accumulated spatial drift and establish an updated origin anchor.

03Directional Vivaldi + compact payloadPassive gain and 240-byte telemetry envelope

An end-fire planar Vivaldi antenna provides a specified +4 to +7 dBi directive gain; compact packets contain target coordinates, altitude, battery status and navigation mode within a 240-byte limit.

04Multi-hop aerial forwardingAdjacent UAVs relay coordinate packets

Narrowband transceivers receive serialized coordinate payloads from adjacent UAVs and re-radiate them toward a ground terminal.

05Autonomous landing clearancePoint-cloud plane fit and alternate-site search

Verify a landing plane against a 15° inclination limit, 0.3 m obstacle threshold and 1.0 m radius; scan an alternate zone within a 5 m radius and restrict final descent to lateral drift below 0.5 m/s.

06Fault-tolerant propulsionSingle-motor failure response

Monitor motor behaviour, alter motor-mix commands to surviving motors and execute return-to-launch upon a rotational-frequency anomaly.

07Bi-directional acoustic relaySurvivor-to-operator and operator-to-survivor audio

Capture and encode local speech, send compressed audio through the wideband link, decode incoming operator audio and disable the audio stream when operating on narrowband radio.

08Ground server + mobile terminalAggregate, de-duplicate and map target locations

A remote ground node synchronises target coordinates across both links, deduplicates common detections and renders a continuous cartographic representation in the mobile interface during satellite outages.

09Independent navigation methodMonitor → anchor → ENU → transform → localise → transmit

A method for GNSS-denied navigation and target localization using synchronized visual and depth streams, thermal verification and directional RF transmission of absolute geographic coordinates.

10Failover + re-engagement methodMove to narrowband, then reset on GNSS recovery

When Wi-Fi RSSI falls below −75 dBm, suspend video and divert coordinates and telemetry to the sub-gigahertz transceiver; after GNSS recovery, blend the fix, clear drift and establish a new anchor.

DOCUMENT BASISSupplied patent specification
TECHNICAL DRAWINGSFigures 1–18 from supplied drawing deck
ADDITIONAL REDRAWSFigures 19–20 reconstructed from text

17 ENGINEERING DRAWING LIBRARY

Twenty figures.
One system architecture.

Browse the supplied patent drawings. Select any figure to enlarge it. Figures 1–18 are original assets extracted from the drawing presentation; Figures 19–20 are clean redraws based on the written specification.

SELECT A DRAWING TO ZOOM ↗

18 TECHNICAL INDEX

System parameters
at a glance.

A compact index of the principal hardware, software, thresholds and performance values described in the supplied specification.

FLIGHTPixhawk 2.4.8 · ArduCopter 4.xEKF3 · DSHOT300 · six 650 KV motors
COMPUTERaspberry Pi 5 · 8 GBQuad-core ARM Cortex-A76 · 2.4 GHz
RGB-DIntel RealSense D435i30 fps capture / 3D depth projection
SLAMORB-SLAM3NEON SIMD · ~25 fps tracking
GNSSRTK / WGS84HDOP >5 trigger · HDOP <2 re-engage
THERMALInfrared signature fusionThermal verification of visual candidates
WIDEBAND2.4 / 5.8 GHz Wi-FiH.264 RTSP + Opus + telemetry
NARROWBANDRYLR998 / SX1276 · 865 MHzSF9 · BW125 kHz · CR 4/5
ANTENNAPlanar Vivaldi / end-fire+4 to +7 dBi specified passive gain
BACKENDFastAPI · MAVLink 2.0WebSocket 2 Hz · GStreamer RTSP
CLIENTAndroid / KotlinGoogle Maps SDK · ExoPlayer · voice
MISSIONSurvey / detect / relay / landAutonomous strip planning and safe touchdown