Autonomy's
Missing Half

At three in the morning on a ridge above 12,000 feet, the load is carried by a person. India holds more than 15,000 km of land border, a 740 km Line of Control and a 3,500 km Line of Actual Control, largely on foot, with roughly 5.15 lakh personnel across the BSF, ITBP, SSB and Assam Rifles. Where the ground refuses wheels — glacier, marsh, scree, rubble, stairwell, tunnel — the Army's answer has been the same since 1947: men, and about six thousand live mules across fifteen Animal Transport units.

India has automated the air above that ground at speed. It has not automated the ground. Its entire fielded legged-robot inventory is one hundred units, ordered once, in September 2023, through an emergency-procurement tranche, with follow-on explicitly contingent on results that have never been publicly reported. Against twelve credible defence mission categories for a quadruped, four have a confirmed Indian programme, three exist only as vendor claims, and five have no Indian activity whatsoever — explosive ordnance disposal, CBRN detection, tunnel and cave clearing, communications relay, and mine and IED detection.

Untu AI is a robotics company building the platform and the autonomy stack for those gaps. Its founding technologist holds a robotics PhD from IIT Jodhpur and has already delivered quadruped navigation and gait planning once before. The company's prototype, Rumi, is not a render and not a rebadged import: the mechanism was designed in code, printed in-house, assembled from locally sourced off-the-shelf parts, simulated, and driven by a controller the team trained itself. In May 2026 it walked — untethered, under joystick command, with every watt and every FLOP on board — ten months after the first CAD file was opened, with no bought chassis, no licensed controller and no black box anywhere in the stack.

Three things converged to make this the year. In May 2025, roughly 70% of the 500-plus drones deployed in Operation Sindoor were disabled by electronic warfare, and by September all 46 indigenous manufacturers put through the Army's GPS-denied trials failed them. In August 2026, Unitree priced its Shanghai listing at $9.04 billion holding 69.75% of the global quadruped market — making the category's dominant supplier a listed champion of the country the platform is meant to face, and therefore unusable under MoD's Trusted Source mandate. And between 2023 and 2026, GPU-parallel simulation and zero-shot sim-to-real transfer collapsed legged control from a national-laboratory problem into a small-team problem. Three years ago, a three-person team could not have done this. That is the "why now".

Untu AI is raising ₹5 crore for 10% across 36 months at a burn of ₹1.3–1.5 crore a year. The round buys the one dependency everything else waits on — quasi-direct-drive actuation — then outdoor terrain, then vision in the control loop, then a trial-grade platform filed against a live problem statement. It sits underneath a ₹50 crore ladder of non-dilutive government capital (iDEX ₹10 Cr, ADITI ₹25 Cr, TDF ₹50 Cr at up to 90% of project cost) and in front of a buyer with ₹1.39 lakh crore of capital acquisition ring-fenced for domestic industry in FY 2026-27. This paper argues that Indian capital should fund it: the ground layer is the unbuilt half of national autonomy, the asset that compounds in legged robotics is the training loop rather than the chassis, and this team has closed that loop on real hardware for a fraction of what the category's funded players have spent.

Autonomy's Missing Half

India has built a serious, funded, working aerial autonomy capability. It has not built the half that touches the ground — and 2025 was the year that omission became operational.

The concession comes first, because it is real. Indian defence innovation has moved fastest in the air. iDEX challenges have funded a generation of drone companies; Project Storm and the Army's AI Roadmap 2025-27 put unmanned aerial systems and swarms at the centre of the modernisation story; the country has domestic airframes, domestic payloads and a credible export posture. Nobody serious argues India cannot build things that fly.

What India has not built is anything that walks. That distinction matters more than it sounds, because the terrain that defines India's security problem is precisely the terrain that refuses wheels and tracks. A wheeled UGV needs ground a wheel can hold. Siachen does not provide it. Neither does the Rann of Kutch, the scree of the LAC, the forested insurgency belt of the northeast, a collapsed building, a staircase inside a cordoned house, or a cross-border infiltration tunnel. Every one of those is a place where a person currently goes first.

The reason is subtle

Legged locomotion was, until very recently, not an engineering discipline but a research frontier held by perhaps a dozen laboratories on earth. Making a machine with four contact points balance, absorb an unplanned impact, and recover from a slip is not a mechanical problem that can be solved by better brackets; it is a control problem, and for thirty years the only people who could solve it worked at Boston Dynamics, ETH Zürich, MIT, or their direct academic descendants. Everyone else bought the result or did without.

India did without. The workaround it pays for instead is men and mules: the Army Service Corps has historically managed on the order of six thousand animals across fifteen units, in continuous use from 1947-48 through Kargil to the present. That workaround is not a quirk. It is the cost — measured in casualty exposure, in tempo, in the number of soldiers whose job is to carry rather than to fight.

Why this matters now

Three independent forces turned a long-standing capability gap into an urgent one, and each arrived in a different domain.

1. Operational — May 2025. Approximately 70% of the 500-plus drones deployed in Operation Sindoor were disabled by electronic warfare. By September, all 46 indigenous manufacturers put through the Army's GPS-denied trials had failed them, and roughly ₹10,000 crore was directed toward EW-hardened prototypes. Autonomy did not fail. India had built only the half that flies — and the half that flies is the half that jams.
2. Structural — August 2026. Unitree Robotics priced its Shanghai STAR Market listing on 6 August 2026 at a $9.04 billion valuation, raising roughly $904 million, holding 69.75% of the global quadruped robot market. The world's default supplier of legged robots is now a listed Chinese national champion. For a platform whose purpose is a China-facing border, that is not a price advantage to be exploited — it is a supply chain that MoD's Trusted Source mandate has already ruled out.
3. Technological — 2023 to 2026. GPU-parallel physics simulation and zero-shot sim-to-real transfer changed who can build a walking robot. Frameworks in the Isaac Lab class run thousands of concurrent training environments entirely on a single GPU, and policies trained that way now transfer to hardware without hand-tuning. What required a national laboratory a decade ago now requires a good engineer, a workstation and a year. The barrier that kept legged robotics out of India was never ambition or capital — it was access to a control capability that has just been democratised.

They are complementary rather than substitutable, and the honest end-state is a quadcopter that scouts and hands off to a quadruped that enters. But India has built one side of that handoff and not the other, and the side it has not built is the side that survives contest.

The prize

Ground autonomy sits upstream of a long list of capabilities that stay manual until it exists. Border patrol and persistent surveillance. Counter-infiltration inside tunnels. Approach and inspection of suspected ordnance. CBRN sampling. Casualty support. Communications relay across mountain terrain where line-of-sight fails. Disaster search inside structures that rescuers cannot safely enter. And beyond defence entirely: industrial inspection, mining, plant and substation patrol — the dual-use markets that ANYbotics has already turned into a business with more than 200 robots shipped and thousands of autonomous inspections a week.

The ambition is a fleet of Indian legged platforms running missions unattended. That is an audacious end-state and it is not what this round funds. The wedge that gets there is concrete: the five documented defence mission categories where no Indian platform exists at all, entered with a machine whose control stack the company actually owns.

A Robot We Own End to End

The Indian category standard is integration. Untu's bet is that the durable asset in legged robotics is the training loop — and that a company which does not own the loop cannot fix what it cannot change.

Define the category by what it is not. The prevailing model for an Indian quadruped programme is assembly: source actuators, source compute, source or adapt a controller, wrap the result in an Indian chassis, and present it as an indigenous platform. That model answers a real question — can we field a robot this year? — and it has produced fielded hardware, which is more than most claimants can say. What it does not answer is the second question: can we change how it moves?

The distinction stopped being academic in September 2025. Forty-six indigenous manufacturers entered the Army's GPS-denied trials and none passed. A failure rate of 46 out of 46 is not a distribution of engineering quality; it is a structural fact about where the autonomy came from. You cannot debug a controller you did not train.

The loop

Untu's stack is a five-stage cycle run repeatedly, and it is the whole method. Design — the mechanism is defined in code, so a change in geometry is a change in a parameter. Simulate — that same definition becomes a physics model accurate enough to train against. Train — a controller is learned across thousands of parallel runs rather than hand-tuned joint by joint. Transfer — the learned controller is loaded onto the robot's onboard computer and run on real hardware. Measure — wherever the machine disagrees with the model, the difference is measured and fed back into the model.

The last stage is the one that is usually skipped and the one that decides everything. System identification — measuring the real machine and correcting the model it trains against — is what turns a simulation into a place where a useful policy can be learned. Untu completed a full identification pass in December 2025, five months before the robot walked.

Two architectural choices

First: the mechanism is parametric. Because the geometry lives in code rather than in a drawing, redesign is a re-run. The observable consequence is that Untu completed two full build cycles — design, print, assemble, instrument, measure — between July 2025 and August 2026, with the second-generation structure built from what the first one taught them. Hardware companies that cannot iterate at that cadence spend their seed round discovering their first mistake.

Second: the controller is learned, not authored. The policy that walks the robot was trained in simulation across thousands of parallel runs and transferred to a Raspberry Pi on the machine's back. No external computer sits in the loop and no tether supplies power. This is a company-reported and hardware-demonstrated result, first achieved in May 2026.

What is running, marked honestly

A memo that blurs simulation and hardware is worthless to a technical diligence process, so the line is drawn explicitly.

The gap between the demonstrated rows and the simulated row is not a software gap. It is the actuator. Rumi runs on low-cost geared servos, which were the correct choice for proving the loop cheaply and are the precise reason it walks at 0.3 m/s on a flat floor and no faster and no rougher. Geared servos are not backdrivable, so impacts go into the gearbox rather than into a compliant leg; their torque density is low; their control bandwidth is limited. The move to quasi-direct-drive actuation is step one of the roadmap rather than step three for exactly this reason — everything downstream depends on it, and none of it requires new research.

The product family, and what compounds

• Rumi. The walking prototype. The proof that the loop closes on metal.
• The field platform. Quasi-direct-drive actuation, outdoor terrain competence, vision in the control loop. This is what the round builds.
• Mission software. Patrol a route, inspect an asset, detect an anomaly, carry a payload between points, and report it in reviewable form. The recurring layer, and the layer that compounds.

Hardware sells the first unit. The terrain and mission policy library is what compounds. Every terrain trained against, every disagreement between the model and the machine measured and fed back, becomes reusable across every future unit and every future customer. This is the ordinary data-and-network advantage of a learned system, and it has a sharp corollary in this market: an integrator's fleet does not get smarter. A fleet running a controller the operator did not train improves only when its supplier ships a new firmware build — and if that supplier is offshore, it improves at the supplier's convenience and on the supplier's terms.

Untu AI: Company Profile

Three founders, one robotics PhD who has solved this problem once before, and a machine that walks.

The pattern that started it

The observation at the centre of the company is that India's robotics ambition keeps arriving as procurement rather than as capability. The country can buy a robot, catalogue it, parade it, and deploy a hundred of them — and still fail a GPS-denied trial forty-six times out of forty-six, because the part that failed was never anyone's to fix. Capability that is purchased cannot be improved; it can only be repurchased.

Missing The founders' own account of the founding moment — the date of incorporation, the trigger, and the decision to build rather than integrate — is not in the source material and should replace this paragraph in the investor-facing version. An origin story lands when it contains a date and a specific event.

The prior work

The technical claim is not speculative. Sudhir Pratap Yadav holds a PhD in robotics from IIT Jodhpur and has previously delivered quadruped navigation using reinforcement learning and terrain-adaptive gait planning, alongside SLAM and planning work, 26-DOF humanoid kinematics at IIIT-Hyderabad, and vision-based manipulation with deep reinforcement learning on a UR5 arm. Prior affiliations include iHub Drishti and TCS. This is a second attempt at a problem the founding technologist has already solved once in an institutional setting — which is a materially different risk profile from a first attempt.

The build, month by month

Traction and validation

• The hardest binary in legged robotics is cleared. The robot walks, untethered, on its own power and its own compute. The overwhelming majority of legged programmes never reach this line; it is what separates a render from a robot, and it is not partially achievable.
• The sim-to-real path closes end to end. Single and dual leg mechanisms track simulated behaviour on the test rig. This is the transfer proof — the evidence that the simulation is a place where useful policies can be learned rather than a rendering of a wish.
• Control risk was retired before hardware risk was taken. In September 2025 the team trained policies and proved them on a commercial quadruped before committing to their own. That sequencing is a judgement signal: it is what an experienced robotics team does and what a first-timer skips.
• Iteration is demonstrated. A full redesign-and-reprint cycle completed and driven by measurement, not by opinion. Two build cycles in thirteen months.
• Capital efficiency. Empty CAD file to walking robot in ten months on a pre-institutional budget. Missing — the actual rupee figure spent to date is not documented, and this bullet is materially stronger with it. It is also the single most persuasive number available to this company: it is the direct comparator against every funded player in Section V.

The roadmap, in dependency order

The sequence below is a dependency chain rather than a preference. Vision cannot rescue a machine that lacks the torque to step onto a rock; navigation is meaningless if locomotion fails on the terrain it crosses; and autonomy is only a mission layer on top of a robot that can already get there. These are also, notably, the steps most often shown first and built last.

India's Ground-Autonomy Moment

This is a bet on two simultaneous shifts — India's forced indigenisation of defence capability, and the democratisation of learned legged control. They are not independent, and Untu sits at their intersection.

The coupling is direct. Indigenisation policy creates a mandated domestic buyer for a capability India does not yet possess; the collapse in the cost of legged control makes it possible for a small Indian team to possess it. Either shift alone produces nothing: a mandate without capability produces rebadged imports, and capability without a mandate produces a research group. Together they produce a company.

From buying platforms to owning designs

India's procurement architecture is explicit about this. Under DAP 2020, acquisitions are ranked by how indigenous they are, and the top of the ladder is not a manufacturing threshold — it is a design-ownership test.

The asymmetry is the whole point. A robot can clear a 50% indigenous-content score on Indian brackets, wiring and machining while its locomotion and control are licensed or copied from abroad — which is precisely why MoD has moved to scrutinising design ownership rather than the bill of materials. At the India AI Impact Summit, organisers pulled a university exhibitor's booth after a Chinese-made Unitree Go2 was presented as an indigenous creation. The claim is now checked.

The rival's trajectory as evidence

China did not win legged robotics on doctrine. It won on manufacturing scale, and the result is a single company — Unitree — holding roughly 70% of the world quadruped market, listed in Shanghai at $9.04 billion in August 2026, shipping over 5,500 humanoids in 2025 with revenue quadrupling to ¥1.7 billion at roughly 60% gross margin and more than 40% of that revenue earned overseas. That trajectory is evidence of inevitability more than it is a threat: the category is going to be large, and it is going to be won by whoever manufactures at scale with a controller they own.

The specific Indian irony is sharper. The largest source of imported actuators and motors for Indian robot dogs is China — the same country the platforms are partly meant to counter on the border. That is the real logic behind the Trusted Source mandate, and it is why a quadruped cannot be treated as secure simply because it carries an Indian brand.

The ambition is funded, not aspirational

The grant ladder above a ₹5 crore round is unusually deep for a hardware company. iDEX Prime provides non-dilutive grants up to ₹10 crore for later-stage prototypes addressing identified capability gaps. ADITI — an iDEX sub-scheme with a ₹750 crore outlay — provides up to ₹25 crore at 50% of the product development budget for deep-tech challenges. The DRDO Technology Development Fund provides up to ₹50 crore at up to 90% of project cost. And the emergency procurement lane exists and is proven for this exact category: it is how the MULE order happened at all, bypassing the standard Make-I/II pipeline entirely.

Can a three-person Greater Noida team sell to the Indian Army?

The objection deserves a direct answer rather than a deflection, and the answer is on the record. The Army's only confirmed quadruped programme was awarded to Aero Arc — a company founded in 2018 that has raised approximately $4.79 million in total, across a single round from a single investor, and subsequently took a ₹40 crore investment from Japan's ACSL. It won a hundred-unit order through emergency procurement from a standing start. The entry ticket in this category is a demonstrated platform against a live problem statement, not balance-sheet scale.

The Competitive Landscape

Untu competes on two fronts — Indian integrators with distribution, and global platform vendors with scale. It fits cleanly into neither camp, and that is the position rather than the weakness.

Versus the Indian incumbents

Concede their strengths in full, because they are real. Aero Arc has the only fielded programme in the country: 100 MULE units ordered in September 2023 under the Army's fourth emergency-procurement tranche, at a media-reported ₹285–300 crore, deployed to the Line of Control at Sunderbani, to Siachen, and into 3 Corps forward-area exercises in Arunachal Pradesh, with deliveries reported complete in April 2026. Zen Technologies is a listed company with a ₹15,551 crore market capitalisation as of 10 August 2026, a launched armed platform in Prahasta, and a 45.33% acquired stake in Bhairav Robotics. Addverb has Reliance behind Trakr. These are companies with real distribution, real balance sheets and real government relationships.

The architectural argument is that every one of them competes on the same two axes: chassis and payload, for two missions — carry and watch — plus a weapons mount. Across twelve credible defence utility categories, four have a confirmed Indian programme (surveillance, logistics, armed platform, and a disaster-response pilot), three exist as aspirational or vendor claims, and five have no Indian activity of any kind. Their autonomy is integrated rather than owned, which is the structural explanation for a 46-out-of-46 failure rate in a GPS-denied trial.

Untu's declared wedge is three of the five empty categories — confined-space entry, hazard approach, and denied-environment autonomy — chosen because they are the ones where the missing capability is control rather than chassis, and therefore the ones an integrator structurally cannot enter.

Versus the global platform vendors

Untu does not out-manufacture any of them, and the memo should not pretend otherwise. Unitree and Deep Robotics own price and volume. Boston Dynamics and ANYbotics own the industrial premium — ANYbotics has raised more than $150 million, shipped over 200 ANYmal units, and is delivering the first Ex-certified legged robot for explosive atmospheres in 2026. Ghost Robotics owns Western defence: LIG Nex1 and Korea Investment Private Equity acquired a 60% controlling stake for $240 million in July 2024, valuing the company at $400 million.

The buyer-side reason Untu can win anyway is that for a platform crossing a contested border, provenance is a specification rather than a preference. A Chinese platform is disqualified by mandate regardless of price. A US or Swiss platform is priced and export-controlled out of a 3,500-kilometre problem — Vision 60 units have been reported at roughly $130,000 each, which does not scale to a border. The addressable position is a platform that is cheap enough for volume, Indian enough for IDDM, and autonomous enough to pass the trial that has so far failed everyone.

Defensible advantages

• Category focus. The five unaddressed missions, not carry-and-watch with a weapons mount. Untu is not trying to out-payload the incumbents; it is entering categories they do not serve.
• Architecture. The loop is owned end to end and the geometry is a parameter, so capability changes are re-runs rather than redesign programmes.
• Proof rather than claim. A machine that walks on its own compute, with sim-to-real closed and marked honestly against what is still simulation.
• A compounding asset. The terrain and mission policy library improves with every deployment and every measured failure; an integrated fleet does not.
• Structural eligibility. Indigenous on the axes IDDM actually scores — design ownership, autonomy, control, comms and encryption, structure — with imports declared rather than disguised.
• Capital efficiency. A walking robot reached pre-institutionally, against peers that have raised $150M (ANYbotics), been valued at $400M (Ghost Robotics) and listed at $9B (Unitree). The ₹5 crore ask sits under a ₹50 crore non-dilutive ladder.

The Investment Case

The market

Global quadruped robot market estimates cluster between $2.4 and $3.7 billion in 2026, scaling to roughly $15 to $17.5 billion by 2035 at a compound growth rate between 16.6% and 18.8%. The spread is a definitional artefact: the wider figures fold in industrial inspection and consumer platforms, the narrower ones count commercial and defence units only. Asia-Pacific is the fastest-growing region and, uncomfortably for this thesis, most of that growth currently accrues to a single Chinese supplier.

India's serviceable position is best framed as a replacement cost rather than a market size, because there is no reliable India-specific quadruped market figure to cite. What exists is the spend the capability displaces: roughly six thousand animals across fifteen Animal Transport units, the manpower cost of 5.15 lakh border personnel walking routes a machine could walk, and the casualty exposure of sending people first into tunnels, rubble and suspected ordnance. Above that sits ₹1.39 lakh crore per year of mandated-domestic capital acquisition, of which legged ground robots currently take a rounding error.

Unit economics and the path to scale

Four revenue mechanisms are contemplated: outright hardware sale to defence and disaster-response buyers; tiered AMC-style contracts scaled by fleet size and feature set; usage-based billing by mission hour or deployment volume; and tailored enterprise pricing for large institutional clients. The hardware sells the platform; the contracted and usage layers are where a legged-robotics business becomes durable, because a fleet in the field generates maintenance, retraining and capability-upgrade revenue for as long as it operates.

The gating process is trials rather than certification. Field evaluation and IDDM categorisation are the 12-to-24-month clock that stands between a working platform and a tranche order, and the purpose of this round is to start that clock rather than to finish it.

What ₹5 crore enables

• Quasi-direct-drive actuation and a second-generation machine. The single dependency every later capability waits on. Behaviours already proven in simulation — faster gaits, jumping, hard landings — are waiting on this change and not on new research.
• Terrain-robust outdoor locomotion. The gate is a full outdoor run, repeated, without a rescue. This is the milestone that moves the platform from a laboratory artefact to something a defence evaluator will look at.
• Vision in the control loop. The gate is a flight of stairs climbed on the first attempt. Stairs are the demonstration that separates a legged robot from an expensive wheeled one.
• A trial-grade platform and the first filing. Two field-ready robots with spares, and eligibility filed against a live problem statement in the first quarter of the round — this company's explicit first milestone, and the answer to its single largest gap.
• Access to the non-dilutive ladder. A demonstrated field platform is the precondition for iDEX, ADITI and TDF applications. ₹5 crore of equity is what unlocks the ability to compete for ₹50 crore that costs no equity at all.

Exit landscape

1. Strategic acquisition — the most natural path. The precedent is recent and directly on point: LIG Nex1 with Korea Investment Private Equity took 60% of Ghost Robotics for $240 million at a $400 million valuation in July 2024, because a defence prime needed to own a proven legged platform rather than partner for one. The Indian analogue already behaves this way — Zen Technologies, at a ₹15,551 crore market capitalisation, acquired 45.33% of Bhairav Robotics in early 2025. Listed Indian defence platforms, large private primes and DPSUs all have the same structural reason to buy an autonomy stack they cannot build.
2. Domestic listing. India's public markets pay real multiples for defence and advanced-manufacturing names, and Zen Technologies is the named, checkable local precedent — a simulation and robotics company at ₹15,551 crore. This is a longer path and requires the recurring layer to be working, but it is not hypothetical.
3. Independent scale-up into dual use. Industrial inspection, mining, plant and substation patrol have shorter sales cycles than defence and do not require a tranche order to generate revenue. ANYbotics has already demonstrated that this is a business rather than a hedge. For Untu this path doubles as the mitigation for monopsony risk.

The capital climate is favourable in a way it has not previously been. Global venture funding into defence, national security and law-enforcement startups passed $14.6 billion in 2026 to date, against a full-year record of $9.6 billion set in 2025. Indian early-stage funds are explicitly allocating to defence as a deep-tech sub-sector. The constraint on Indian defence robotics has not been investor appetite; it has been the absence of companies with a demonstrated platform to fund.

Risks and Mitigants

Stated in the hardest form an investor would put them. Three founders, a research prototype walking at 0.3 m/s on a flat indoor floor, no customer contact, and the slowest-moving buyer in the country. The mitigants below are facts already in motion, not reassurance.

Why Indian Capital, and Why Now

The opportunity in plain terms

India has the engineering talent and a government writing ₹1.39 lakh crore of mandated-domestic cheques a year. What it has rarely had is capital willing to fund a hardware team through the years when the robot is slow, ugly and indoors. That is the period this company is in, and it is the only period during which the price is low.

This is not a software startup with a hockey-stick slide. It is three people, a printed robot, a controller they trained themselves, and a walk that has already happened. It has no revenue, no letter of intent and no customer meeting on the record — and it has cleared the one binary in legged robotics that cannot be faked, marketed around, or partially achieved.

The window

The window is the gap between an emergency tranche and a programme of record. India ordered one hundred ground robots in September 2023 with follow-on explicitly contingent on results that have never been reported. No named budget line for legged UGVs exists. No CDS, Army Chief or Vice Chief statement on quadrupeds for border security has been made publicly. No Indian counter-quadruped doctrine exists despite documented Chinese demonstrations. Meanwhile the category's dominant supplier listed at $9 billion this month, global defence-tech venture funding set a record before the year was two-thirds done, and the Army's own GPS-denied trial established a published, testable discriminator that nobody in India has yet passed.

A capability with no programme of record, a mandated buyer, a published performance discriminator, and no domestic company that owns its control stack is not a crowded market. It is an opening, and openings in defence procurement close when somebody writes the requirement.

Why this thesis

1. The ground layer is the unbuilt half of national autonomy, and 2025 demonstrated operationally that the built half does not survive contest.
2. The durable asset in legged robotics is the training loop, not the chassis — and the 46-of-46 GPS-denied failure is the clearest evidence available that Indian platforms integrated their autonomy instead of owning it.
3. This founder has solved this problem once already, in an institutional setting, and has now closed the sim-to-real loop a second time on hardware the team built themselves.

The ask is unglamorous and specific: fund the actuators, get the robot outdoors, put sight in the control loop, and file against a live problem statement inside the first quarter.

Information gaps in this document

Everything below was searched for in the company's own material and not found. Items marked external were resolved by research and are already reflected in the text; items marked internal can only be supplied by the founders and are flagged inline where they appear.