ONTIC LABS
The foundation for physical intelligence
DECK - NOVEMBER 2026
ONTIC LABS
AI conquered the digital world.
It still can't fold a shirt.
By 2032, anyone can give any robot a new task,
JUST by talking to it.
ONTIC LABS
VISION
3
Teaching a robot as easy as briefing a colleague.
Robust robot learning: >99% success in new scenarios.
ONTIC LABS
GOAL
4
Three ways to teach robots today.
Classic simulators
✕ CAN'T CAPTURE THE WORLD'S COMPLEXITY AND SIM2REAL GAP
√ TRAIN TO PERFECTION
ONTIC LABS
THREE PARADIGMS
5
Three ways to teach robots today.
Classic simulators
✕ CAN'T CAPTURE THE WORLD'S COMPLEXITY AND SIM2REAL GAP
√ TRAIN TO PERFECTION
Teleoperation and VLAs
✕ DOESN'T SCALE: HUNDREDS OF HOURS PER TASK
✕ HARD TO ACHIEVE 100% SUCCESS
ONTIC LABS
THREE PARADIGMS
6
Three ways to teach robots today.
Classic simulators
✕ CAN'T CAPTURE THE WORLD'S COMPLEXITY AND SIM2REAL GAP
√ TRAIN TO PERFECTION
Teleoperation and VLAs
✕ DOESN'T SCALE: HUNDREDS OF HOURS PER TASK
✕ HARD TO ACHIEVE 100% SUCCESS
Video world models
✕ VIOLATES PHYSICS
ONTIC LABS
THREE PARADIGMS
7
The bottleneck isn't just scale or data.
It's a missing ingredient.
ONTIC LABS
OUR BET
8
Video world models learn pixels, not physics.
SAME INTERACTION, SIX DIFFERENT PICTURES
SAME INTERACTION, ONE 3D STATE
ONTIC LABS
WHAT IS MISSING
9
3D particles make physical state learnable.
ONTIC LABS
OUR SOLUTION
10
From ordinary video to a complete 3D world.
WE RECONSTRUCT THE WHOLE SCENE, INCLUDING THE PARTS THE CAMERA NEVER SAW
ONTIC LABS
PROGRESS
11
ONTIC LABS
Track everything in 3D over time.
Now scaling to large video.
Tracking in 4d from videos
12
PROGRESS
ONTIC LABS
Hands, objects and robots, reconstructed in 3D.
Now scaling to large video.
RECONSTRUCTED HANDS & OBJECTS
13
PROGRESS
Interact in the simulated world.
ONTIC LABS
WORLD MODEL
14
A robot skill learned with our representation
ONTIC LABS
SELF IMPROVING POLICIES
15
68% → 94%
success after the policy improves itself
Robots rehearse and learn tasks in our world model.
ONTIC LABS
ROBOT LEARNING
16
Built SOTA models for human and 3D scene reconstruction
Prof. Gerard Pons-Moll
University of Tübingen, >15y
perception
robotics
interaction
ONTIC LABS
TEAM
17
Built SOTA models for human and 3D scene reconstruction
Prof. Gerard Pons-Moll
University of Tübingen, >15y
3D humans & contact
perception
robotics
interaction
ONTIC LABS
TEAM
18
perception
robotics
interaction
Prof. Gerard Pons-Moll
Dr. Wieland Brendel
University of Tübingen
3D humans & contact
Max-Planck & ELLIS Institute Tübingen
Co-Founder Maddox.AI & Polybot
Scaling & robustness
ONTIC LABS
TEAM
19
perception
robotics
Prof. Gerard Pons-Moll
Dr. Wieland Brendel
University of Tübingen
3D humans & contact
Max-Planck & ELLIS Institute Tübingen
Co-Founder Maddox.AI & Polybot
Scaling & robustness
Edgar Sucar
University of Oxford (VGG)
3D/4D scene understanding
ONTIC LABS
TEAM
20
perception
robotics
Prof. Gerard Pons-Moll
Dr. Wieland Brendel
University of Tübingen
3D humans & contact
Max-Planck & ELLIS Institute Tübingen
Co-Founder Maddox.AI & Polybot
Scaling & robustness
Edgar Sucar
University of Oxford (VGG)
3D/4D scene understanding
Mikel Zhobro
University of Tübingen
World modelling
ONTIC LABS
TEAM
21
perception
robotics
Prof. Gerard Pons-Moll
Dr. Wieland Brendel
University of Tübingen
3D humans & contact
Max-Planck & ELLIS Institute Tübingen
Co-Founder Maddox.AI & Polybot
Scaling & robustness
Edgar Sucar
University of Oxford (VGG)
3D/4D scene understanding
Mikel Zhobro
University of Tübingen
World modelling
Omid Taheri
Max Planck Institute for
Intelligent Systems & Adobe
Human-object interaction
ONTIC LABS
TEAM
22
perception
robotics
Prof. Gerard Pons-Moll
Dr. Wieland Brendel
University of Tübingen
3D humans & contact
Max-Planck & ELLIS Institute Tübingen
Co-Founder Maddox.AI & Polybot
Scaling & robustness
Edgar Sucar
University of Oxford (VGG)
3D/4D scene understanding
Mikel Zhobro
University of Tübingen
World modelling
Omid Taheri
Max Planck Institute for
Intelligent Systems & Adobe
Human-object interaction
Prof. Georg Martius
University of Tübingen & Max-Planck
Co-Founder Polybot
Robot learning, model-based RL
ONTIC LABS
TEAM
23
perception
robotics
Prof. Gerard Pons-Moll
Dr. Wieland Brendel
University of Tübingen
3D humans & contact
Max-Planck & ELLIS Institute Tübingen
Co-Founder Maddox.AI & Polybot
Scaling & robustness
Edgar Sucar
University of Oxford (VGG)
3D/4D scene understanding
Mikel Zhobro
University of Tübingen
World modelling
Omid Taheri
Max Planck Institute for
Intelligent Systems & Adobe
Human-object interaction
Prof. Georg Martius
University of Tübingen & Max-Planck
Co-Founder Polybot
Robot learning, model-based RL
ONTIC LABS
TEAM
24
perception
robotics
Prof. Gerard Pons-Moll
Dr. Wieland Brendel
University of Tübingen
3D humans & contact
Max-Planck & ELLIS Institute Tübingen
Co-Founder Maddox.AI & Polybot
Scaling & robustness
Edgar Sucar
University of Oxford (VGG)
3D/4D scene understanding
Mikel Zhobro
University of Tübingen
World modelling
Omid Taheri
Max Planck Institute for
Intelligent Systems & Adobe
Human-object interaction
Prof. Georg Martius
University of Tübingen & Max-Planck
Co-Founder Polybot
Robot learning, model-based RL
ONTIC LABS
TEAM
25
Wonbong
Giorgio
Boya
4D Vision, Data
& Robot Learning
Shuk
Pradyumna
perception
robotics
Prof. Gerard Pons-Moll
Dr. Wieland Brendel
University of Tübingen
3D humans & contact
Max-Planck & ELLIS Institute Tübingen
Co-Founder Maddox.AI & Polybot
Scaling & robustness
Edgar Sucar
University of Oxford (VGG)
3D/4D scene understanding
Mikel Zhobro
University of Tübingen
World modelling
Omid Taheri
Max Planck Institute for
Intelligent Systems & Adobe
Human-object interaction
Prof. Georg Martius
University of Tübingen & Max-Planck
Co-Founder Polybot
Robot learning, model-based RL
Wonbong
Giorgio
Boya
4D Vision, Data
& Robot Learning
Shuk
Pradyumna
ONTIC LABS
TEAM
26
The robots are here. The intelligence is missing.
$50B
spent on robots in 2025
ABI RESEARCH, 2025
13 Million
Humanoids at work by 2035. 1 billion at work by 2050.
MORGAN STANLEY, APR 2025
ONTIC LABS
MARKET
27
We start where robots already work.
WAREHOUSE AUTOMATION, GLOBAL
$30B
2025
$63B
2030
Market size, US dollars · Mordor Intelligence, 2025
Logistics + parcel
sorting · depalletising · truck loading
Warehouses + fulfilment
picking · packing · kitting
Retail + grocery
shelf restocking · backroom handling
Factories
assembly · machine tending · line changeover
ONTIC LABS
MARKET
28
ONTIC LABS
LANDSCAPE
We compete on the intelligence.
We work with makers.
ROBOT FOUNDATION MODELS
Physical Intelligence · Skild AI · Generalist · Field AI
Learn from teleoperated data, collected by hand
VIDEO AND LATENT WORLD MODELS
Runway · Decart · Odyssey · AMI Labs · General Intuition
Learn pixels, not physics
3D WORLDS AND SIMULATION
World Labs · Genesis AI · Lightwheel · Hillbot
Hand-built or static worlds, little real interaction
ONTIC
3D physical state, learned from real video, built for robots.
ROBOT MAKERS AND INTEGRATORS
NEURA Robotics (Partnership)
Agile Robots
Bosch (Partnership)
Wandelbots
Astribot
Unitree
Poke & Wiggle (Partnership)
29
€30M for 24 months:
scaling the world model to paying fleets.
ONTIC LABS
THE ASK
30
€ MILLION LEFT FROM THE ROUND
€0M
€10M
€20M
€30M
0
6
12
18
24
Model v1
3 robot types
API alpha
Paid fleets
~1,000-GPU run
~1,000-GPU run
Months after the round closes. Ends with a €1.5M reserve.
TEAM
15 → 35 people
COMPUTE
~5M GPU-hours + own cluster
ROBOT TYPES
2 → 3+
CUSTOMERS
Partners → paid fleets
RUNWAY
24 months + €1.5M reserve
Mainly on compute and talent.
ONTIC LABS
COST PLAN
31
WHAT WE SPEND EACH QUARTER
Compute
People
Robots + data
Ops
€0M
€2M
€4M
€6M
€4.1M
Q1
€2.4M
Q2
€2.5M
Q3
€5.2M
Q4
€2.9M
Q5
€5.5M
Q6
€2.9M
Q7
€2.9M
Q8
Own servers
Run 1
Run 2
Quarter after the round closes
Where the €30M goes.
ONTIC LABS
USE OF FUNDS - €30M
32
€13.7M
Compute · 46%
Reserved cloud GPUs, always on
128, then 160 GPUs · 2.5M GPU-h
€6.3M
Two training runs
1,024 GPUs × 6 weeks · 2.1M GPU-h
€5.2M
Own B300 cluster
32 more GPUs, plus hosting
€2.2M
€8.6M
People · 29%
Salaries, 15 → 35 people
broken down on the next two slides
€8.1M
Recruiting and relocation
search fees for senior roles
€0.5M
€3.8M
Robots + data · 13%
Video data
licensing, capture, labelling
€2.4M
Robots and lab
third robot type, cells, sensors
€1.4M
€3.9M
Ops + reserve · 13%
Legal, IP, finance, software
€50k a month
€1.2M
Space, Tübingen lab
€25k a month
€0.6M
Pilots and travel
mostly in year 2
€0.6M
Reserve
5% of the round
€1.5M
Cloud GPUs at €2.50 per GPU-hour. Salaries are loaded costs. Anything raised above €30M goes into compute.
ONTIC
From the Greek ontos – true existence.
OUR MISSION
Model the world as it really is.
33