Purpose of RoboOI

Train with biology.
Deploy on silicon.

We use living human neurons to teach robots to adapt with fewer real-world trials, then ship what they learn as ordinary software.

→ How it works →
↓

½

of a robot policy's success can be lost to a 3 cm camera shift. arXiv ↗

~800k

living human neurons on each CL1 culture we train with. IEEE ↗

0M

industrial robots in operation worldwide. IFR ↗

How it works

Neurons learn.
Silicon ships.

  1. 01

    Living human neurons learn your task Run on real cultures

    We encode the task as stimulation on a CL1 culture of lab-grown human neurons, and it learns in a closed loop.

  2. 02

    Our GPUs distil it

    We compress what changed into a control prior and benchmark it against SAC and PPO.

  3. 03

    Your robot runs it

    The prior ships as ordinary software for your controller. No neurons leave the lab.

01Living neuronsLearn the task in closed loop
0264-electrode arrayStimulate and record
03Control priorWhat changed, distilled
04Silicon controllerWhat the robot runs

Fig. 1 · The RoboOI transfer stack. Point at a layer to inspect it.

Pre-registered Hard controls

CH 01–64 · Culture

Sim

Spike rate

0.0ev/s

Study design · Experiment 01

Pre-registered

Does a culture-trained prior cut real robot interactions?

Culture
1 neuron culture, closed loop
Task
1 robot skill, held-out evaluation
Arms
Trained · untrained · shuffled · from scratch
Baselines
SAC and PPO at matched budgets
Metric
Interactions to 80% success

Measured,
not promised.

Pre-registered controls and strong silicon baselines. We publish the result either way.

For robotics teams

What a pilot
looks like

One painful task, a measured answer either way.

  1. 01

    You bring a task

    One skill that breaks when conditions shift, plus your sim and logs.

    Week 0
  2. 02

    We run the benchmark

    Our culture-trained prior against strong silicon baselines, pre-registered.

    Weeks 1–6
  3. 03

    You get the result

    A validated prior if it wins. A full report either way.

    Deliverable
↓

Questions

Straight answers

What we do, what's proven, and what isn't yet.

01What is organoid intelligence?

Computing with living, lab-grown human brain tissue, such as neuron cultures grown on multi-electrode arrays. Johns Hopkins researchers named the field in 2023.

02Do RoboOI robots run on living neurons?

No. The neurons are a training instrument and stay in the lab. Your robot runs a conventional controller on its own hardware.

03Does it work yet?

We've run our pipeline on real human neuron cultures. Whether the prior makes a robot learn faster is what our pre-registered experiment measures, and we'll publish the result either way.

04Who is RoboOI for?

Robotics teams whose policies break when a camera, surface, gripper or part changes, plus labs and investors following the work.

Early access

Join the
waitlist.

A small number of pilots, robotics teams first. Tell us what breaks.

  • Robotics teamsAn adaptation audit of your policy, then a bio-trained prior pilot on one task.
  • LabsShared benchmarks and encoding tools for cultures on multi-electrode arrays.
  • InvestorsUpdates on the transfer proof as results come in.

We only use this to reply to you.