JUNO
Cognitive Reconnaissance Officer.
One operator. One AI. The whole picture.
She commands the mission, not the vehicle. She is a persistent intelligence: she has been watching for six hours before you walk in. By the time you sit down, she already knows the safe corridors, the dwell points, the anomalies. She briefs you in a sentence. You decide. She acts.
Feed · Live
Grid 06 · Clear
Dwell 06:14:22
She has been watching before you arrived.
The world model is built continuously from live telemetry, sensor feeds, and environmental data. When you brief yourself, the picture is already synthesised — you do not start from zero.
She speaks to you in plain language.
No HUD to parse, no data panel to cross-reference. She delivers the operational brief as a sentence and takes commands as a sentence. The interface is conversation.
She commands the mission, not the vehicle.
The operator issues intent — “hold the northern perimeter, flag any heat signature that enters grid six.” She translates that into validated instructions the fleet executes. You remain at the mission layer.
She generates the mission report automatically.
Time-stamped event logs, anomaly flags, operator confirmations — written throughout the mission. The debrief document exists the moment the mission ends. Nothing is reconstructed from memory.
One brain. Three bodies.
The same Juno core — sensing, map fusion, voice, reasoning — ships in three chassis. What changes is the fleet around it and the body it lives in. Three markets, one company.
“You write Juno once — the sensing, the map fusion, the voice, the reasoning. The three products are the identical core wrapped in different amounts of fleet. That is how you reach three markets without building three companies.”
Juno Mobile
On the go — the unit that comes to the site.
- Fleet — 4 drones + 2 rovers, deployed from the vehicle.
- Range — local; wherever the vehicle drives. Tactical, fast, self-contained.
- Compute — a rugged edge box (Orin / RTX) in the vehicle.
- For — survey crews, construction sites, search & rescue, mobile mine units, rapid response.
Juno Station
The fixed regional base — it owns an area, persistently.
- Fleet — 4 long-range + 10 mid-range + 20 short-range: a real coverage pyramid.
- Range — wide; long-range eyes hold the perimeter, mids do the work, shorts handle detail and indoor.
- Compute — where a full DGX / server rack belongs: fixed building, mains power, proper cooling.
- For — large mines, ports, refineries, solar & wind farms, border and critical-site monitoring, estates.
Juno Maritime
For the ship — the offshore unit. The hardest, highest-value one.
- Fleet — marinised drones + a small USV (surface) and/or ROV (underwater) in place of rovers.
- Range — vessel-relative; launch and recover from a moving deck.
- Compute — a rugged server in the ship; the ship’s power is available.
- For — offshore oil & gas, shipping & port authorities, offshore wind, coast guard, marine research.
Same brain. Growing fleets. Three chassis.
| — | Juno Mobile | Juno Station | Juno Maritime |
|---|---|---|---|
| Body | Vehicle | Building | Ship |
| Fleet | 4 drones + 2 rovers | 4 + 10 + 20 drones | Marine drones + USV / ROV |
| Range | Local | Regional | Vessel-relative |
| Compute | Rugged edge box | Full DGX / rack | Rugged ship server |
| Same brain? | Juno ● | Juno ● | Juno ● |
Build Mobile first. Cheapest to build, it is the demo, and it sells to surveyors and mining fast.
Station is the big-revenue product. Same software — just more fleet and a proper server. Sell it once Mobile has proven the brain.
Maritime last. Highest value per unit, but the hardest to build and slowest to sell. Earn the right to build it.
You write Juno once — the sensing, the map fusion, the voice, the reasoning.
The three products are just different amounts of fleet and a different chassis around the identical core. That is how you reach three markets — survey, infrastructure, offshore — without building three companies.
Running today. End to end.
A full pipeline — natural-language command, reasoning, fleet control and a spoken brief — runs today, entirely on-device. Everything below is working code.
- Full pipeline: natural-language command → LLM reasoning → drone response → voice brief, end-to-end in simulation
- In-process simulated fleet at 2 Hz telemetry — position, heading, battery state updated continuously
- Complete state machine: idle, patrolling, investigating, returning, emergency
- Battery drain simulation with automatic return-to-launch (auto-RTL) below threshold
- GPU-gated voice output — degrades gracefully to text on CPU-only hardware, no crash
- YOLO detection path wired in — simulated object-detection events fed into the world model
- Operator confirmation gate enforced — she never dispatches without an explicit confirmation token
- Mission log written to disk throughout — time-stamped, event-tagged
Indian military data never leaves the box.
No foreign server in the loop. No cloud API call. No model checkpoint hosted abroad. The entire reasoning stack — LLM, world model, voice synthesis, telemetry processing — runs on hardware in the operator’s physical custody.
“Air-gap capable is not a feature. It is the architecture. Every design decision was made assuming the network does not exist and the adversary is listening. Atmanirbhar Bharat is not a slogan here — it is the product specification.”
No dependency on foreign infrastructure. Cloud AI products — even the best — route data through servers in foreign jurisdictions. For defence applications, that is not a compliance issue. It is a disqualifier. Juno is designed from the ground up to be self-contained.
Bought once, owned permanently. No subscription. No API key that can be revoked. No model provider that can change its terms. Once deployed, the system belongs to the operator — permanently, irrevocably.
The model trains and runs locally. The local LLM checkpoint is stored on-device. Updates are applied offline. The system functions identically with no network connection — by design, not by fallback.
Data sovereignty is non-negotiable. Intelligence generated by the system — flight paths, contact logs, operator decisions — never leaves the deployment environment. There is no telemetry home. There is no usage analytics. The operator’s data is the operator’s.
She recommends. You decide. Always.
“She prepares the picture. She recommends the course of action. She never acts unilaterally on anything with lethal potential — not at v0.1, not at v2.0, not at any version of this system. The confirmation gate is architectural, not a setting. It cannot be toggled off.”
Monitoring, synthesis, preparation.
She watches. She builds the world model. She identifies anomalies, calculates routes, drafts instructions, and assembles the brief. All of this happens without operator input — it is her job between confirmations.
Anything with consequence in the physical world.
No drone moves without a confirmed instruction from the operator. No reconnaissance tasking, no repositioning, nothing that touches the physical world executes without an explicit operator confirmation token. This is enforced at the code level.
The trust is not that she will do the right thing. The trust is that she will not do anything until you tell her to.
Where we are. Where we are going.
A clear line from a working system today, to a field-ready prototype, to programme scale.
Foundation
- Full NL → LLM → drone → voice pipeline in simulation
- In-process fleet simulation at 2 Hz telemetry
- Complete state machine (5 states)
- Battery drain + auto-RTL
- GPU-gated voice, graceful CPU fallback
- YOLO detection path wired in
- Operator confirmation gate enforced
- Mission log written continuously
- MAVSDK hardware-control abstraction layer — code-complete
- Hardware: RTX 3060 development rig
Field-Ready Prototype
- MAVSDK real-hardware validation — first physical flight
- Multi-drone coordination (3-drone fleet)
- Narrative mission report generation
- Holographic command presence via Looking Glass display
- Real sensor fusion (camera + optional LiDAR)
- Field-grade UI hardening and latency budgeting
- iDEX DISC application and demo readiness
- Hardware path: Jetson Orin edge compute
Programme Scale
- Extended human-in-the-loop doctrine — formal verification of confirmation gate
- Multi-operator coordination layer
- Persistent area intelligence across mission cycles
- Red-team adversarial robustness testing
- Integration pathways for BEL/HAL programme vehicles
- Field kit: hardened enclosure, satellite-fallback comms
Six hours of watching. One safe corridor. Someone comes home.
A soldier is trapped behind enemy lines. There is no extraction route visible on the map. There has been no route for six hours. But she has been watching the terrain for those six hours — the patrol patterns, the gaps, the windows. She already knows the corridor. She has already drafted the fleet’s tasking. When you sit down, she says: “There is a route. Do you want me to send them?” You say yes. The fleet goes. Someone’s son comes home.
She does not win the battle. She gives the operator the picture, the plan, and the half-second that changes the outcome.
The channels built for this.
India has constructed a defence innovation ecosystem specifically to develop sovereign capability of this kind. Juno is designed to fit the evaluation criteria and procurement pathways of each of the following.
iDEX DISC
- Defence India Startup Challenge (DISC) is the primary early-stage entry point
- Problem statements in autonomous systems and AI command support are published each cycle
- Structured grant funding through the published DISC rounds
- Path to trials and service evaluation follows successful DISC completion
IISc · Academic Partnership
- Collaborative research on human-machine teaming in autonomous systems
- Access to computational resources and domain expertise for field testing
- Academic publication pathway — critical for credibility in defence procurement
- Fellowship and project funding pathways available
TDF / DRDO
- Technology Development Fund supports dual-use technology with defence applications
- DRDO co-development agreements available for promising indigenous capability
- Grant and in-kind support — test facilities and datasets
- Path to DRDO Young Scientist Laboratory (DYSL) collaboration possible
Defence Primes
- BEL and HAL are the natural system-integration partners at programme scale
- L&T Defence for ruggedised hardware procurement and field kit development
- Primes provide the procurement pathway into larger force-level programmes
- Engagement follows iDEX validation, on programme terms
“The ecosystem is real, the pathways are open, and the product fits the criteria. iDEX, TDF/DRDO and the defence primes are purpose-built to fund exactly this kind of sovereign capability.”
“Leave the world in a better place than you found it.”