← Back
untu.ai · platform · 2026
The chassis is a commodity. The walk is not.

We are not a robot company.
We are the layer between.

Untu AI builds the design→simulate→train→transfer loop that turns any legged mechanism into a walking one. Rumi is our proof. The loop is the product.

Raising ₹5 Cr · platform build-out
01
The provocation

Anyone can weld a robot dog.
Almost nobody can make it walk.

Every legged-robot project on earth reinvents the same unsolved software problem: a controller that turns a pile of actuators into coordinated motion. The chassis — motors, links, a frame — is a solved, shoppable commodity. The part that decides where the foot goes is not.

$ ls ./legged-robotics-stack/
chassis.cad      # commodity — buy it
actuators.bom  # commodity — buy it
compute.spec   # commodity — buy it
controller/     # unsolved — this is the company
Untu AI · The Layer Between
02
Not vaporware — we run it ourselves first

Rumi walks. Untethered, on its own compute.

0.3 m/s
Walking, joystick-commanded, May 2026
10 mo
Empty CAD file to hardware walk
2
Full mechanism iterations on the same loop
Untu AI · The Layer Between
03
The architecture

One loop. Any chassis you plug into it.

01 design
Mechanism as code — geometry is a parameter
→
02 simulate
Physics twin, GPU-parallel
→
03 train
Policy learned across 1000s of parallel runs
→
04 transfer
Runs on any onboard compute target
→
05 measure
Real-world gap fed back into the sim

Rumi's chassis is one instantiation of the "design" node. Swap the chassis, the loop still runs — that portability is the entire thesis, and it's why we built our own hardware first: to prove the loop generalises, not just that we can build one robot.

Untu AI · The Layer Between
04
The product, unbundled

Three layers. Sell all three, separately.

SDK

Sim-to-real pipeline

The design→simulate→train→transfer tooling, packaged for a partner's own chassis and compute target.

Policy library

Trained terrain & gait behaviours

Walk, stand, recover, climb — pretrained policies a partner fine-tunes onto their own hardware instead of starting from zero.

Reference platform

Rumi

Proof the stack works end to end, and a fielded product in its own right — the flagship customer of our own platform.

Hardware sells the first unit. The policy library compounds — every terrain trained and every failure measured is reusable across every future partner and every future chassis.

Untu AI · The Layer Between
05
Why this is priced like software

Autonomy stacks get software multiples.
Chassis get hardware multiples.

The market has already priced this distinction: portable autonomy software has repeatedly traded on a software multiple rather than a defence-prime hardware multiple in recent rounds — the gap between an integrator and a stack owner shows up directly in valuation, not just in capability.

Untu AI · The Layer Between
06
Go-to-market

Prove it on Rumi. License it to everyone building a chassis.

StageWhat happensRevenue shape
NowRumi walks untethered — the loop is proven on our own hardwarePre-revenue, proof stage
This roundPackage SDK + policy library; onboard 2–3 hardware partners (Indian integrators, industrial OEMs) who own chassis but not autonomyLicensing pilots
Year 2Partners ship units running our stack; each deployed unit reports terrain data back into the shared policy libraryPer-unit royalty
Year 3+The policy library is the moat — more deployed partner units than any single competitor could train aloneCompounding, capital-light
Untu AI · The Layer Between
07
The objection, answered

"Why would a hardware team buy this instead of building it?"

What building it themselves costs
  • Legged control was, until recently, a research capability held by a dozen labs on Earth
  • Most hardware-first teams stall exactly here — a chassis with no controller that can walk it
  • 0 of 46 indigenous Indian makers passed the Sept 2025 GPS-denied trial — the failure is the software, not the metal
What licensing costs instead
  • A proven loop, already validated end to end on real hardware — not a research bet
  • A policy library that starts wider than any single team's training data
  • Royalty economics instead of a multi-year in-house robotics-research build-out
Untu AI · The Layer Between
08
The ask

Fund the platform layer.
Not another robot.

₹5 Cr packages the SDK and policy library, and signs the first hardware-partner licensing pilots — capital-light, higher-margin, and it scales without us building more chassis ourselves.

₹5 Cr · 10% · platform milestones
Untu AI · The Layer Between
09