Zero-code AI robotics

Humans teach
the work.
Robots learn it.

Stop programming robots. Start showing them what to do. A worker demonstrates the task with our grippers — Ludis turns it into training data, and the robot learns to handle the real-world variation. Built on the Universal Manipulation Interface.

handheld UMI grippers
fisheye GoPros
6-DoF fiducial pose tracking
0 teleop rigs or robots to start
Two Ludis handheld UMI grippers in use
LD-04R · CAPTURE 00:42:17
DICT 4×4 · 2 TAGS LOCKED
≥95%
of raw data usable
3 cams
2 wrist + 1 chest
6-DoF
pose per gripper
30s
demos — many, not few

Before you commit

Is Ludis right for you?

Honest adoption starts with hard questions — about your task, your reliability bar, the economics, and us. Here's the checklist we'd work through with you. A fit is proven in a pilot, not a pitch.

07 Is my task a good match for this paradigm?
  • Is it genuine manipulation — grasping, placing, inserting, folding — rather than something a simpler machine could do?
  • How repetitive vs. variable is it? Learned policies tolerate some variation but get shaky with high variability or many edge cases.
  • What precision and tolerance does it need? Sub-millimeter assembly is a different beast than "drop item in bin."
  • Are the objects awkward — deformable, transparent, reflective, fragile, heavy? Those are classically hard.
  • Is the work environment controlled, or messy and unpredictable?
  • How long is the task horizon? A single short action is far easier than a long multi-step sequence.
  • Can the policy hit your required cycle time, or only do it slowly?
03 What reliability do I actually need — and can it clear that bar?
  • What success rate is acceptable — 95%? 99.9%? Many learned systems plateau below industrial-grade reliability for critical steps.
  • What is the cost of a failure — scrap, safety, downstream defects?
  • How are exceptions and recovery handled? Is there a human fallback when it gets confused?
03 Is the data / demonstration side feasible for me?
  • Who collects the demonstrations, how many are needed, and how good do they have to be?
  • When my product or task changes, do I have to re-demonstrate everything — and how painful is that?
  • Do I need in-house ML/robotics talent to keep it healthy, or is it turnkey?
03 Does the economics work?
  • Total cost: hardware, integration, licensing, ongoing maintenance — not just the sticker price.
  • What labor or throughput am I actually displacing, and what is the payback period?
  • How does it compare to my real alternatives: fixed automation, a conventionally programmed arm, or keeping humans?
03 How does it fit my operation?
  • Footprint, power, safety guarding, and how it slots into my existing line / workflow.
  • Uptime, support, SLAs — and who is on the hook when it breaks.
  • Update path when tasks evolve.
03 Is Ludis itself a safe bet?
  • How mature are they — funding, track record, real deployed customers in my industry?
  • Since UMI is published research, what is their actual proprietary value-add and moat?
  • Lock-in risk: what happens to my line if they fold or get acquired?
02 Safety, compliance, and people
  • Human-robot safety, plus any regulatory constraints (food, pharma, medical, etc.).
  • Workforce impact, change management, and acceptance.
01 Can I de-risk before committing?
  • Is there a narrow, low-stakes pilot task I can run first — to get real data before betting the operation on it?

No fabricated benchmarks here — bring these questions and we'll answer them against your task, with a low-stakes pilot to get real numbers first.

The problem

Traditional automation can't keep up with real work.

Conventional robots are hard-coded for one rigid motion. Reality isn't rigid — parts move, deform, and arrive out of place. So the work stays manual.

Hard-coded robots

  • ×Hundreds of engineering hours to program one task
  • ×Breaks when a part shifts by a single millimeter
  • ×One cell, one task — no tolerance for variation
  • ×Every product change means re-programming

Show, don't program

  • A worker demonstrates the task in minutes
  • Learns to handle misaligned, moving, messy parts
  • The same skill transfers across the fleet
  • New tasks mean more demos — not more code

How it works

Teach a robot the way you'd teach a person — by showing it.

No code, no teleoperation rig, no robot on day one — and no line downtime. Workers train just 1–2 hours a day while normal production continues.

01 / 03

Demonstrate

Pick up the grippers and do the task by hand — packing, sorting, assembling. They move like your own hands, so there's nothing to learn.

02 / 03

Record

Two wrist fisheye cameras — with side mirrors for implicit stereo — plus a chest camera capture the scene. ArUco markers track each gripper in full 6-DoF. No motion-capture stage required.

03 / 03

Automate

Upload the session. A diffusion policy learns the task — reacting to misaligned and moving parts — then the same gripper runs it on a standard robot arm, autonomously.

See the transfer

One demonstration. Then the robot does it.

Record a person doing the task once. The same gripper — mounted on a standard robot arm — reproduces it autonomously, handling the variation a hard-coded program never could.

HUMAN · DEMONSTRATION
DROP CLIP →
ROBOT · AUTONOMOUS
DROP CLIP →
Detail of a Ludis gripper with fiducial markers
GRIPPER · LD-04R

The device

A data-collection tool, not a robot.

A

Twin grippers

3D-printed jaws that mirror a parallel robot gripper — what you do maps one-to-one to what the robot will do. ~85 mm stroke, recovered straight from the markers.

B

Three fisheye cameras

Two GoPros at the wrists (155° Max Lens, 2.7K/60) with side mirrors for implicit stereo, plus a chest GoPro for egocentric scene context.

C

Fiducial pose

ArUco DICT_4×4 markers recover precise 6-DoF gripper pose and jaw width frame by frame — identity and position in one mark.

D

Just record

Battery-powered, untethered, audio-synced across cameras. Capture on the line, in the warehouse, anywhere the real work happens.

The engine

From demonstration to dataset, automatically.

Every session runs through a visual-inertial pipeline that reconstructs exactly what your hands did — in metric 3D, frame by frame.

00

Capture

3 synced cameras

01

Map

SLAM workspace

02

Localize

6-DoF trajectory

03

Detect

fiducial + jaw width

04

Plan

build dataset

05

Train

diffusion policy

≥95%of raw data usable — successful SLAM, zero demos dropped
Built on the Universal Manipulation Interface (Stanford) · ORB-SLAM3 visual-inertial odometry

Built on proven research

Universal Manipulation Interface Diffusion Policy ORB-SLAM3 VIO ArUco fiducials
Backed by LOGO LOGO LOGO

The data flywheel

Labor becomes data. Data becomes autonomy.

Conventional robot programming doesn't scale past one cell. Demonstrations do — and they get cheaper every shift.

Capture

Operators record demonstrations during normal shifts.

Train

Sessions upload and feed a continuously improving policy.

Deploy

Trained skills run on autonomous robots on the floor.

Compound

Every new site and task makes the whole fleet smarter.

Where it works

Start with the tasks robots can't be programmed to do.

If a person can show it in a few minutes, Ludis can capture it.

MESSY BIN → KIT

Dynamic kitting & sorting

Sorting heterogeneous components — valves, bolts, fittings — from messy bins into assembly kits. The policy handles flipped and stacked parts.

SEALS · O-RINGS

Flexible materials handling

Manipulating rubber seals, O-rings, and gaskets that deform as you grip them. Traditional robots struggle here; demonstration-trained policies thrive.

DUAL-ARM ROUTING

Wire harnessing & routing

Using both grippers to pick up cables and pneumatic hoses and route them into clips and connectors — dexterous, two-handed work.

Deployment

Your data isn't locked to our hardware.

The gripper that records is the gripper that runs. Because capture and deployment share the same end-effector, what the human did maps one-to-one onto the robot — on whatever platform fits your floor.

A Immediate ROI

Industrial cobot arms

Mount the gripper on the proven arms you already trust — Universal Robots, xArm, and similar. The policy runs on hardware your team knows.

B Where work moves

Mobile manipulation

Need to cover ground? Place the same arms on standard AGVs to bring trained skills to the line, the bin, or the bench.

C On the roadmap

Humanoids

Full humanoid hands are a frontier — transfer error is still high. We start you on reliable arms today and explore humanoids as the research matures.

Data & security

Your operations, your data, your model.

The dataset you build is a strategic asset — so it stays yours. We make capture safe to run inside a working facility.

A

You own the dataset

Every demonstration you record belongs to you. Export the raw data anytime — no lock-in.

B

You own the policy

The trained model is yours to deploy, retrain, and keep running on your own hardware.

C

On-prem & air-gapped

Capture and process on-site for sensitive facilities. Nothing has to leave the building.

D

Encrypted end to end

Footage is encrypted in transit and at rest, with role-based access controls and audit logs.

The math

What does one manual task cost you?

Estimate the labor tied up in a single repetitive handling task today — time you could redeploy to work robots can't do.

Annual labor on this one task
$192,000
6,000 operator-hours / year · 250 working days
That's the cost of a single repetitive task, today. Ludis lets you capture it once and redeploy that skilled time to higher-value work.

Start a pilot

From first demo to deployed in about 60 days.

A pilot proves Ludis on one of your tasks — measured, low-risk, and without disrupting the line.

Phase 01 · Week 1

Equip & onboard

We fit your operators with grippers and set capture up right on your line.

Phase 02 · Weeks 2–4

Demonstrate

Your team records the target task 1–2 hours a day. Normal production never stops.

Phase 03 · Weeks 4–6

Train

We process the sessions and train the policy off-site on the demonstration data.

Phase 04 · Weeks 6–8

Deploy

The policy runs on your cobot arm. You keep the dataset and the trained model.

The team

Robotics, machine learning, and the factory floor.

Replace these placeholders with your founders and key hires — photos, names, and one-line bios.

PHOTO

Co-founder & CEO

Add name
PHOTO

Co-founder & CTO

Add name
PHOTO

Head of Machine Learning

Add name
PHOTO

Field Engineering Lead

Add name

FAQ

Questions, answered.

What kinds of tasks can Ludis learn?
Repetitive, dexterous handling — kitting and sorting from messy bins, manipulating deformable parts like seals and gaskets, routing cables and hoses. If a person can show it in a few minutes, Ludis can capture it.
How many demonstrations do we need?
It depends on the task's variability, but most pilots collect demonstrations over a few weeks at 1–2 hours a day. More demonstrations of the edge cases make the policy more robust.
Does it work with the robots we already have?
Yes. The gripper that records is the gripper that runs — mount it on standard cobot arms like Universal Robots or xArm, or on AGVs for mobility. Your data isn't tied to our hardware.
Does collecting data interrupt production?
No. Operators record during normal shifts with handheld grippers — there's no teleoperation rig and no robot on the line during capture, so the line keeps moving.
Who owns the data and the trained model?
You do. The dataset and the resulting policy are yours to export, retain, and deploy. On-prem and air-gapped capture are available for sensitive sites.
How long until it's running autonomously?
A typical pilot goes from first demonstration to a policy deployed on your arm in roughly 60 days.
Is my task a good fit for Ludis?
Ludis suits repetitive, dexterous handling with real-world variation — kitting and sorting from messy bins, manipulating deformable parts, routing cables and hoses. It is a weaker fit for sub-millimeter rigid assembly or work that fixed automation already does well. The honest way to know is a short pilot on your actual task.
What happens when our product or task changes?
New or changed tasks mean more demonstrations, not re-programming. You record the new variation and the policy learns it — incremental, not a rebuild.
Do we need in-house ML or robotics talent?
No. Operators collect demonstrations during normal shifts; Ludis processes the sessions and trains the policy. You keep the dataset and the trained model.

Teach once.
Robotize the rest.

Bring Ludis to your floor and start building a demonstration dataset this quarter.

Book a demo →