Understand the scene, not just the image.
Vision, depth, touch, and proprioception fused into a shared representation of objects, affordances, and people in a space.
Now out of stealth
Teaching machines to seereasonworkcare in the real world.
Wittgenstein Labs builds physical intelligence: AI that gives machines the ability to perceive, reason about, and act in the physical world, safely and usefully, alongside people.
Mission
Language models learned to describe the world. The next frontier is intelligence that can live in it: that understands weight, friction, and fragility, and turns knowledge into the careful work of human hands.
The limits of a machine's language have long been the limits of its world. We are building the models, the data, and the deployments that push those limits out into kitchens, hospitals, workshops, and factory floors.
ApproachFrom pixels
From pixels
to purpose.
Physical intelligence is a closed loop. Every deployed robot sees more, every task teaches the model, and every improvement returns to the floor.
A world model that anticipates.
Models that predict how things move, break, spill, and respond, so a machine can plan the next action before it takes it.
Dexterous, safe, embodied control.
Policies that translate intent into precise motion on real hardware, with the margins of safety that shared human spaces demand.
Every task makes the next one easier.
Demonstrations and deployments flow back into training, compounding into skills that transfer across tasks, sites, and embodiments.
What we're building
Three bets on the physical world.
We work across the full stack of physical intelligence, from the knowledge layer of an organization to the data that trains robots to the machines that do the work.
Data
Collecting and selling robot training data.
Robots learn from examples of real work. We capture high-quality, multimodal demonstrations of physical tasks and package them as training data for teams building embodied AI.
Everyday manual tasksPhysical activities people do with their hands every day.
CookingCleaningRepairing equipmentOther manual activities
Hospital logisticsMoving supplies, specimens, and equipment through clinical environments.
Surgical intelligenceData and models for precise, high-stakes manipulation in the operating room.
Robot caregiving dataAssistive tasks that support patients and caregivers with dignity and safety.
Knowledge
A company brain.
The know-how of an operation lives in people, procedures, and machines. Our company brain captures that working knowledge, connects it to what is happening on the ground, and makes it available to every person and robot in the organization.
People & procedures
Sensors & systems
Teams & robots
Deployment
Deploying trained robots into factories.
Research only matters when it survives contact with reality. We take trained robots onto factory floors, integrate them with existing lines and teams, and keep improving them from what they learn on the job.
IntegrateFit into existing lines, workflows, and safety systems.
OperateRun real shifts alongside the people who know the floor.
ImproveFeed every task back into the next generation of skills.
How we work
Grounded in research
Rooted in the physical intelligence community at Harvard and MIT, and committed to rigorous, measurable progress.
Built for the real world
We measure success on real tasks in real environments, not on benchmarks alone.
Safe by design
Robots that share space with people must be predictable, transparent, and careful. We treat that as a first principle.
News
Latest from the lab.
Announcements, research notes, and updates from Wittgenstein Labs.
Announcement
Wittgenstein Labs emerges from stealth to build intelligence for the physical world
Incubated at Company Ventures' Terrarium arm and led by Harvard-MIT experts in physical intelligence.
Read the announcement →Coming soon
Research notes from the lab.
We'll share what we learn as we build: data, models, and lessons from the field.
Careers & partnerships
Help machines
Help machines
understand the world.
We're looking for researchers, engineers, and operators who want to bring AI into the physical world, and for partners with real tasks to solve.