Cameras can’t feel.
Robots now learn by trial and error, and cameras grade every try. But a camera can’t tell a gentle grip from a crushing one. pors.ai can.
See the productTalk to the foundersThe quality-control layer for robot learning. Public test live. Product in build.
A camera can’t tell gentle from crushing.
Drag the grip. The picture barely changes until the egg is already broken.
Goods get damaged.
Robot hands wear out early.
Robots learn rough habits.
Why now: robots are teaching themselves.
Before
Engineers programmed every move by hand.
Now
Robots practise thousands of times a day.
The bottleneck
Every try needs a grade. People can’t keep up, so cameras do the grading.
The data to fix it doesn’t exist.
We tested it. Camera grading breaks.
Share of attempts a camera-based grader judged correctly. From our open-source test, robot-eval-coverage.
Our fix: sensors in test objects, not in the robot.
Today
A person watches video and guesses how hard the robot gripped.
With pors.ai
The object records the true grip, every time. Customers change nothing on their robots.
How we make money.
Sensor kits
Sold near cost. They put us on customers’ robots and collect the data.
Grading service
Paid per hour of robot practice graded. Grows with every robot deployed.
The dataset
Licensed to robot AI builders. This is the lasting asset.
Customers: robot AI labs, factories, and gripper makers.
Methods can be copied in months. Our data can’t.
It only exists because someone physically collected it.
Milestones, each with a pass-or-stop test.
Done
Open-source test shows camera grading fails when the grip is hidden.
Weeks 1–6
Prove touch data beats video alone on robots it has never seen.
Months 2–5
Record 10,000 attempts across three robot types.
Months 5–9
Build the grader. It must beat the best camera-only one.
Months 9–14
Show robots trained with it succeed as often and break far less.
Talk to the founders.
swaraj@porsai.comServices.
Work with us before the product ships.
Sensor kits
Test objects that record grip strength, synced to your cameras.
Grip audit
How hard your robots really squeeze, by robot and shift.
Grading pilot
Our grades next to yours, on the same tasks.
Custom data
Recordings of your task, with grip strength included.
Test design
Success, force and damage reported together.
How it runs
Scope
One task, one question.
Install
Fit the sensors and sync them.
Grade
Record attempts and score them.
Report
Findings, plus your data.
Priced per project. You keep your data.
Do we need new sensors on our robots?
No. The sensors are in the test objects.
What if touch data doesn’t help?
We’ll tell you.
Tell us about your robots.
Request a pilotpors.ai Judge.
One grader, two products.
Grading API
Send a robot’s attempt, get a grade back, including how hard it gripped.
Force Monitor
A factory dashboard showing grip strength over time, with alerts.
An example grade
{
"succeeded": true,
"task_progress": 0.92,
"peak_grip_force_n": 31.4,
"damage_events": 1,
"damage_started_at_frame": 212
}
Illustrative. A camera-only grader would score this a pass and miss the damage.
What it works with
| Source | Why |
|---|---|
| Our test objects | Main route. No changes to the robot. |
| Motor current | Most robot arms already report it. |
| Wrist force sensor | Most accurate. |
| Fingertip touch | Catches slipping. |
| Video only | Always works. |
Public test live. Product in build.
Be an early partner.
Request accessGiving robots a sense of touch.
The idea
Robots learn by practising.
Practice needs a grade.
Every grade comes from video.
We build the grader that can feel.
The team
A scientist builds the grader. An entrepreneur with corporate experience builds the data.
Snigdha Chandan
Founder, CTO
A scientist who has done research across several companies. Builds the grader: the model that learns from video and grip strength together.
Swaraj Dhar
Co-founder, CTO
An entrepreneur who has built a business and worked inside large companies. Builds the data: test objects, recording and the dataset.