Backed byYCombinator

Sento builds sensorized everyday tools that turn the work people already do into training data for robots.

Language models learned from trillions of tokens on the internet. Robots have no internet to learn from. They need physical interaction data, and today it is slow, expensive and hard to scale.

Today’s methods trade fidelity for scale.

ScalePhysical fidelityTeleoperation-gradeVideo scale

Teleoperation

People remotely control robots. The data is high quality, but expensive and limited by the robot hardware available.

Egocentric video

Head-mounted cameras scale easily, but miss the forces, contact and precise trajectories that manipulation depends on.

Wearable sensing

Gloves and suits capture richer signals, but get in the way of natural work and are hard to deploy.

Sento

Sensors built into the tools people already use. Teleoperation-grade signals, collected at the scale of video.

Scroll to plot each approach.

Instrument the tool, not the worker.

Teleoperation needs a robot. Video can’t feel force. Wearables get in the way. Sento puts the sensors where the work actually happens: in the tool.

01

An ordinary tool, rebuilt with sensors.

Sento makes inexpensive, sensorized versions of everyday tools: screwdrivers, hammers, wrenches. A sensing skin around the grip maps three-axis force wherever the hand touches.

Every tool has a signature.

Each Sento tool records precise 6-DoF motion, a three-axis force map across the whole grip, torque about every axis, contact, vibration and temperature, synchronized with first-person video. Driving a screw, hammering a nail, tightening a bolt: every task leaves its own trace.

The internet-scale dataset robotics is missing.

Our long-term goal is a distributed network of instrumented tools in factories, workshops and warehouses, continuously generating large, diverse, physically grounded manipulation data.

Built for frontier robotics teams training general-purpose manipulation systems and vision-language-action models.

  • Pretraining

    Broad physical interaction data across tools, tasks and environments for robotic foundation models.

  • Imitation learning

    Demonstrations of skilled human technique, with the forces and contacts that video can’t see.

  • Post-training

    Targeted data for the specific skills and settings a policy still needs to master.

Book a call with the founders.

If you’re training manipulation models, pick a time and we’ll walk you through the data.

  • What a Sento tool records, and how
  • Datasets, and how to get access
  • The tasks and tools your robots need to learn

Prefer email? founders@sentorobotics.com