Daksha AI Drones — Autonomous Search & Rescue Drones

01 — The problem

Too much ground, too little time.

Survival collapses fast

Search speed is the single biggest determinant of how many people are found alive.

The network dies first

Towers and power fail early, so cloud-dependent and remotely-piloted drones lose their link exactly when they are needed most.

A tired operator misses people

Drones stream video to a human, and hours of thermal feed will hide the person sitting in the frame.

Coverage cannot scale

Trained pilots are scarce and one pilot flies one aircraft, so coverage does not grow with the disaster.

The gap is not aerial access — agencies already own drones. The gap is that those drones cannot search, decide or report without a human and a network behind them.

10 hrs

Disasters scatter survivors across terrain too large, too unstable and too dark for ground teams to sweep inside this window.

1 : 5

Ratio of trained pilots to drones today. One pilot flies one aircraft — manpower is the bottleneck, not hardware.

02 — Proposed solution

A self-flying rescue aircraft, no cloud required.

The entire perception and navigation stack runs on a Qualcomm edge processor — no cloud, no ground link required.

▶ Try the live detection console 🎬 Watch the mission reel The console runs the detection stack live in your browser; the reel walks the full flight sequence. Both need an access code.

01 · AUTONOMY

On-board autonomy, zero connectivity

The Qualcomm QRB5165 / RB5 platform runs detection, SLAM and path planning on the Hexagon NPU inside the airframe — no ground station, no LTE, no cloud.

02 · SENSING

Multimodal victim detection

Fused RGB, long-wave thermal and acoustic sensing finds people under debris, under canopy and in darkness, including partial occlusion where single-sensor models fail.

03 · PLANNING

Self-planned search coverage

Given a polygon, the drone generates and continuously re-plans its coverage pattern, avoiding obstacles from onboard depth and SLAM rather than a pre-surveyed map.

04 · IMAGING

Thermal and visual as primary signal

Long-wave thermal picks up body heat through darkness, smoke and light debris; the visual camera confirms context and rejects false positives — one pass, day and night.

05 · OUTPUT

Commander-ready output

Not raw video: geo-tagged victim pins with confidence scores, triage priority and a live 3D site map, pushed to the incident commander the moment a drone re-enters mesh range.

03 — What makes it different

Built for the moment the network drops.

Compared on	Conventional SAR drone	Daksha
Connectivity	Needs a continuous RF or LTE link to a ground station	Flies the full mission offline; mesh is opportunistic, not required
Who finds the victim	A human operator watching a live video feed	On-board multimodal AI — the operator supervises rather than searches
Scaling	One pilot flies one aircraft	One operator supervises an N-drone swarm
Compute	Cloud inference or a GPU ground station	Edge NPU under 15 W, with no impact on endurance
What comes back	Hours of raw video to review	Geo-tagged victim map with confidence and triage priority

Defensibility — detection under partial occlusion and acoustic distress cues require field-collected thermal data that does not exist publicly. The dataset we build becomes the moat.

04 — Target customers

Three government buyers, one fast lane.

PRIMARY

NDRF & State Disaster Response Forces

16 NDRF battalions and SDRFs across 28 states are the mandated first responders for earthquake, flood and landslide search, procuring through NDMA and state budgets.

SECONDARY

Armed forces & paramilitary

Army, ITBP and BSF run high-altitude and avalanche rescue where ground search is slowest and connectivity is nil — the sharpest fit for offline autonomy.

TERTIARY

Power, oil, gas & mining

Remote plants, pipelines and mines carry statutory emergency-response duties. Private buyers, no tender, far shorter procurement cycles.

ADJACENT

Coast Guard & forest departments

Maritime SAR for the Coast Guard, plus lost-trekker and wildfire response for state forest departments. Same airframe, retrained models.

Three of four are government buyers: long cycles, but high contract value, low churn and a strong reference effect once the first agency deploys. The industrial segment buys privately — the fastest route to first revenue.

05 — Market opportunity

₹4,200 Cr addressable in India alone.

TAM

₹43,000 Cr

Global public-safety, emergency-response and SAR drone market by 2030 (~USD 5.2 B), across hardware, payloads and mission software.

SAM

₹4,200 Cr

India-addressable slice: disaster-response and public-safety drone systems procurable by NDRF, SDRFs, paramilitary and municipal services.

SOM

₹525 Cr

Realistic 3-year capture: NDRF plus 6–8 state SDRFs and two paramilitary programmes, counting units, retrofit kits and subscriptions.

NDMA modernisation grants and state disaster-response funds now explicitly budget for aerial search capability.

Drone Rules 2021 and the PLI scheme make Indian-manufactured airframes the procurement default.

Every agency that already owns drones is a retrofit customer — the AI payload sells without replacing the fleet.

Figures are directional top-down estimates for grant review, to be validated with primary procurement quotes during the pilot phase.

06 — Business model

Hardware opens the door; subscriptions keep it open.

Hardware wins the first order, the retrofit kit widens the funnel, and subscription plus support turn each deployment into recurring revenue.

Complete drone units

Integrated airframe with the Qualcomm compute-and-sensor payload, sold per unit to agencies. Unit price and margin target [TBD].

ONE-TIME

AI retrofit kit

Compute module, thermal and audio sensors as a bolt-on for drones agencies already own — larger installed base, shorter sales cycle.

ONE-TIME

Mission software subscription

Per drone, per year: model updates for new terrain and disaster types, mission planning, mapping and the commander dashboard.

RECURRING

Training & certification

Operator and supervisor courses for agency crews, priced per batch — attaches to nearly every unit sale as procurement norms tighten.

REPEAT

AMC, spares & support

Post-warranty annual maintenance at a typical 12–18% of hardware value, plus spares — the annuity that compounds as the fleet grows.

RECURRING

Two streams are recurring: every unit sold pulls an annual subscription and an AMC behind it, so revenue compounds with the deployed fleet. Pricing marked [TBD] until the first agency quotes.

07 — Go-to-market

Four phases, thirty-six months.

Months 0–6 · Prove

Field pilot, third-party evidence

Field pilot with one SDRF and one NDRF battalion under an MoU. Certified detection metrics captured in a live mock drill, so the evidence is third-party rather than a lab demo.

Months 6–12 · Certify & list

Clear the compliance gate

DGCA type certification, BIS and MeitY compliance, and a GeM listing so agencies can procure directly without an open tender.

Months 12–24 · Land

Sell into state budgets

Sell into 6–8 state SDRFs through state disaster-management budgets and NDMA schemes. Lead with the retrofit kit for the fastest entry.

Months 24–36 · Expand

Paramilitary, municipal, annuity

Move into paramilitary and municipal fire services, and convert every pilot into an AMC plus subscription annuity.

ChannelDirect government sales, supported by system integrators already empanelled with MHA and state agencies. Credibility anchored on the SIH and Qualcomm association plus published field-trial results with a partner institution.

08 — How we use the grant

Every rupee goes to hardware, trials, certification.

Total grant ask

₹10,00,000

Hardware and compute

Qualcomm dev kits, airframes, LWIR thermal camera, depth and audio sensors, batteries

₹3,00,000

Field trials and travel

Drill-site access, agency demos, transport and logistics for on-site testing

₹2,00,000

Certification and compliance

DGCA type certification, testing-lab fees, BIS and airworthiness documentation

₹1,50,000

Data collection and annotation

Building the occluded-victim thermal dataset and paying for labelling

₹1,50,000

Team stipends

6 members over [TBD] months of full-time build

₹50,000

IP and contingency

Provisional patent filing plus a buffer for component price movement

₹1,50,000

Milestone at exitOne certified airframe, a validated detection model with measured recall on real drill footage, and a signed pilot MoU with a state disaster-response agency — deployed over a [TBD] month build-to-field-trial cycle.

09 — Team combination

Six roles, each owning one layer of the stack.

Arpan and Aaditya lead the build end to end — airframe and on-board autonomy. The rest of the team carries research, integration, outreach and delivery.

Team lead

Team lead & hardware

Arpan Sachan

Leads the team and owns the airframe: component selection, sensor mounting, power and assembly.

Core build

Software & edge compute

Aaditya Bhatnagar

Builds the detection software and ports it to the Qualcomm NPU for real-time onboard inference.

Research & development

Sumit Chauhan

Researches thermal and visual detection methods and validates accuracy on real field data.

Implementation

Muskan

Integrates autonomy, path planning and the ground interface, and runs flight testing.

Marketing

Tashu Nauhwar

Owns agency outreach, demos and pilot MoUs with NDRF and state disaster-response forces.

Management

Pratham Upadhyay

Runs schedule, budget and grant milestones, plus certification and compliance paperwork.

10 — Why this matters

7,903 Still counting

lives lost to natural forces in India, 2024 (NCRB)

— one life lost every 66 minutes

2,825

Lightning

1,832

Heat

3,246

Search-and-rescue events

Only the search-and-rescue share is reachable by aerial search. That is the share Daksha is built for. Without the grant, the project stops at a slide deck. With it: a certified, field-proven unit ready for a first state order within 12 months.