NVIDIA Inception Program memberNVIDIA Inception Program member
Silicon ValleyLatin America

We turn the skill in Latin America’s hands into robots that work here.

mano captures real work across Latin American industry, from first-person video to wearable rigs, teleoperation and robots on the floor, and turns it into the data robots learn from. It starts in Mexico.

NVIDIA Inception Program member
Ciudad de México
19.43° N, 99.13° W
San Francisco
37.77° N, 122.42° W
Scroll to explore

Physical AI is starved for real-world data. We capture it here, and build the foundation for robots across Latin America.

FIG 01 — MANO / CAPTURE → TRAIN → DEPLOYPHASE 01 · CAPTURE12345678910ABCDEFGHREC · EGO CAM 01EPOCH --/12BIN ABIN Bt0t1t2t3t4t5ISHOULDER · 2 DOFIIWORKCELL390FIG. 01PICK / PLACEPHASE 01 CAPTUREUNITS mm

Robots learn by watching people work.

  1. ( 01 )

    Physical AI is moving out of the lab and onto factory floors, into kitchens and warehouses. Robots learn those jobs the way people do: by watching real work, done by real hands.

  2. ( 02 )

    Foundation models and robot hardware are becoming widely available. Proprietary data from real operations, captured with ever richer rigs and processed into training-ready form, is the part that can't be downloaded.

  3. ( 03 )

    That work happens every day across Latin American industry, in a range of environments few labs can reach. mano starts with the hands doing it and climbs, rung by rung, to the robot that will.

See how mano works

How mano works

Data collection runs today. Training and deployment are the next layers of the stack.

01
Layer / 01Live now

Data Collection

We bring capture kits onto real floors and record the manual work that already happens there, with motion data and quality control on every clip. Operations keep running while we record, and each rig adds richer, more robot-native data.

  • Egocentric videoLive nowHead-mounted cameras on real work.
  • Handheld grippersNextUMI-style grippers that demonstrate through a robot's own fingers.
  • Wearable rigsNextGloves and backpacks for finger pose, grip force and full-body motion.
  • TeleoperationNextOperators driving real robots through real tasks.
  • Robots on the floorThenDeployed robots recording their own work, and every correction.
02
Layer / 02Next

Model Training

Next, we will fine-tune open-source robot foundation models on that data, tuned to a company's own tasks, objects and environments. Because the models are open, the data keeps its value each time a stronger model is released.

03
Layer / 03Then

Deployment

Then, a trained robot: those models running reliably on robot hardware that already exists, from arms to mobile systems, inside the operation where the data was captured.

Factories and light industryAgricultureWorkshops and repairKitchens and food prepService workHomesControlled capture sites

Anyone can mount a camera for someone else. We own every step after it.

Footage alone doesn’t train a robot. Every mano clip runs through the same pipeline, from raw capture to quality control, de-identification, annotation and segmentation, until it is training-ready.

Loading capture…

Real mano-capture footage with its own QC, hand keypoints, task segments and action labels. Hover a stage to see its layer.

  1. ( 01 )

    Capture

    Ultrawide video, motion data at 100 Hz and lens calibration, recorded during real work. No audio, ever.

  2. ( 02 )

    Sync

    Video and motion aligned to the frame, so every clip carries how the camera and the hands moved.

  3. ( 03 )

    Quality control

    Every half second is judged for camera, hands and task: dark, blurred or blocked frames, missing hands, idle or walking time. Each clip gets a health score and its usable seconds.

  4. ( 04 )

    De-identification

    Logos, screens, documents and bystanders masked by a reviewer frame by frame, with location and identities stripped before anything leaves mano.

  5. ( 05 )

    Annotation

    Hand keypoints, 21 per hand, with the wearer's hands told apart from everyone else's.

  6. ( 06 )

    Segmentation

    Every clip split into the tasks it contains, each with its start, its end and whether it succeeded.

  7. ( 07 )

    Labels

    Every action labelled in plain language: which hand, what it did and to which object, checked for consistency.

  8. ( 08 )

    Export

    Training-ready episodes in standard robotics formats, synced with the motion data.

Open models level the field. Proprietary data tilts it.

Capturing now
01Record real work
02Own a proprietary dataset
03Add richer rigs
04Robots recording on your floor
05A robot trained on your work
Selling data
01Record real work
02Sell the footage once
03The buyer owns the dataset
04Their models learn your work
05Rent the robot back
Waiting
01No footage recorded
02No footage recorded
03Same models as everyone
04Start capturing late
05Catching up

Latin America has always lent its hands. Now it trains the robots.

( 01 )

The work is already here.

Factories, workshops, kitchens and farms across the region run real manual work every day, in more kinds of environments than any lab can stage.

( 02 )

The capture is happening everywhere.

Real work is already being recorded at scale around the world. The robots will be trained either way. The question is whether Latin America takes part as a partner or watches from the side.

( 03 )

From supplier to builder.

For decades the region was treated as a place to send work to. mano brings the technology here, so local industry helps shape it and the region builds its own data and robotics talent.

Who we work with

( 01 )

Physical-AI labs

Real-work data collected to your spec, from environments you can't reach from a lab: egocentric video today, gripper, wearable and teleoperation data as the stack grows.

( 02 )

Industrial companies

Turn the work your teams already do into a proprietary dataset, and a head start on the robots that will learn it. Nothing about your operation has to change.

( 03 )

Universities

Research alliances built on real-world data from Latin American industry, and a path for the region's engineers and researchers to work at the frontier of physical AI, not follow it.

Start with the data you already have.

Labs, industrial companies and universities: tell us what you are building and we will show you where mano fits.

rodrigo@manorobotics.comWhatsApp · +1 (650) 405-9035
I am

Send opens an email to rodrigo@manorobotics.com in your mail app with these details filled in.