India automated the sky. It never built the layer that walks — the one electronic warfare cannot switch off. We did.
15,000+ km of land border. A 740 km Line of Control and a 3,500 km Line of Actual Control, much of it glacier, marsh, desert or forest. 5.15 lakh personnel walk it, because nothing else can.
The largest unautomated frontier in Indian defence is not a technology gap. It is the ground itself.
~70% of 500+ drones in Operation Sindoor disabled by electronic warfare. 0 of 46 indigenous makers passed the Army's GPS-denied trials that followed.
Unitree priced its Shanghai IPO at $9.04B, holding ~70% of the global quadruped market — the dominant supplier is now a listed Chinese national champion.
GPU-parallel simulation collapsed legged control from a national-lab problem to a small-team problem. What needed Boston Dynamics a decade ago, a three-person team can now do in ten months.
| Dimension | Air layer (quadcopter) | Ground layer (quadruped) |
|---|---|---|
| Endurance | Minutes to ~1 hr, battery-limited | Hours, refuel/swap on the move |
| Signature | Audible, radar/EO visible against sky | Low, blends with ground clutter |
| Counter-measures | Jammed, spoofed, shot down | No RF dependency to jam |
| Weather | Grounded by wind, storm | Operates through it |
| Reach | Cannot enter buildings, tunnels, caves | Goes anywhere a person goes |
Air and ground are complementary, not substitutes — but India built only one of the two, and 2025 proved which one fails first.
The incumbent Indian answer is integration: buy actuators, buy compute, buy or copy a controller, wrap it in an Indian paint scheme. That answers "can we field a robot?" — not "can we change how it moves?"
Mechanism defined in code. Geometry is a parameter, not a redraw.
The design becomes a physics model accurate enough to train against.
A controller learned across thousands of parallel runs, not hand-tuned.
Loaded onto onboard compute, run on the real machine — measured, fed back.
Nothing in the stack is rented. Structure printed in-house, parts locally sourced off-the-shelf, tooling open source, controller trained by us. Every layer can be changed — which is exactly what the roadmap requires and an integrator cannot do.
This is a research prototype that has cleared the hardest binary in legged robotics — does it walk on real hardware, on its own. Most legged programmes never reach this line.
FY26-27 capital acquisition ₹1.85 lakh Cr; ~75% ring-fenced for domestic industry.
iDEX up to ₹10 Cr · ADITI up to ₹25 Cr · TDF up to ₹50 Cr at 90% of project cost — non-dilutive, parallel to this round.
Buy (Indian-IDDM) requires 50% indigenous content plus Indian design ownership — the axis MoD now actually scrutinises.
| Player | Origin | Focus | Owns the autonomy stack? / IDDM-eligible? |
|---|---|---|---|
| Untu AI | India | Ground-layer autonomy, five unaddressed missions | Owned end to end · Eligible |
| Aero Arc (MULE) | India | Logistics carry, imported actuators | Integrated · Eligible, first mover |
| Zen Technologies | India | Chassis + payload, ₹15,551 Cr market cap | Integrated · Eligible |
| Unitree | China | Price leader, 70% global share, $9.04B IPO | Owned · Disqualified by mandate |
| Ghost Robotics | USA / LIG Nex1 | Western defence, $400M valuation | Owned · Priced / export-controlled out |
Indian integrators compete on chassis and payload for two missions — carry and watch. None addresses the five documented capability gaps. That is why 0 of 46 passed.
Move off geared servos — a machine that can absorb impact and be pushed hard. The single dependency everything else waits on.
A full outdoor run, repeated, without a rescue. Off the lab floor.
Cameras and depth sensing. A flight of stairs, climbed on the first attempt.
Eligibility against a live problem statement, Q1 of year two.
iDEX/ADITI/TDF non-dilutive funding on the back of a demonstrated platform.
| Risk | Severity | Mitigant |
|---|---|---|
| Single-buyer, slow procurement | High | Non-dilutive grant ladder runs in parallel; MULE itself was bought via emergency procurement from a standing start |
| Prototype-to-field gap | High | The gap is actuation, not research — behaviours already proven in sim are waiting on hardware |
| Better-capitalised rivals | Medium | None compete on the five documented gaps; provenance disqualifies the cheapest supplier |
| No LoI yet | Medium | First milestone of this round: file against a live problem statement |
| Three-founder team, hardware programme | Medium | 50% of use-of-funds is manpower; burn costed against named roles |
India is writing ₹1.39 lakh crore of cheques to buy Indian. It has not yet found an Indian company that can teach a machine to walk.