Why 2025 is a pivotal year for robotics
Anabelle Nicoud explores robotics with Anto and CosmicBrain, including the training data and skills behind useful robots.
By Anabelle NicoudRead the story
Opens on IBM Think in a new tab.Robots in the world.
One operating system.
Scroll to discoverPhysical AI. Put to work.
Multi-brand Robots as a Service. Live model evaluations and benchmarks. Real deployment and teleoperation data.
Plan a deployment01 / One operating layer. Three ways to build.
Deploy. Operate. Learn from the world.
We partner with more than 15 robotics brands to deploy and operate robots. Today, we’re serving hospitality and deploying into data centers, with cloud kitchens and other verticals under exploration.
For physical AI model teams, the same deployment network becomes a place to run live evaluations and benchmarks, and a source of real deployment data and hundreds of thousands of hours of teleoperation data.
With curiosity,The CosmicBrain teamSan Francisco · Planet Earth
For end customersRobots as a Service
Robots, integrated into your everyday operations.
For model teamsEvals, benchmarks & dataTest your models in the field. Learn from real operations.
For robot manufacturersHardware partnershipsYou build the robot. We bring it into operation.
observe → learn → help
02 / The robot deployment operating system
From experience to intelligence.
One operating layer across hardware brands, sites, and models. CosmicBrain connects robot selection, site integration, human operations, live evaluation, and data into a continuous deployment loop.
human experience
+ physical intelligence
a world of possibility
+ physical intelligence
a world of possibility
From a defined task to an operated robot.
Select hardware from our robotics brand partners, integrate it into the site, and support the workflow with human supervision and teleoperation. Robot as a Service brings those pieces into one operating relationship.
01Match the hardware
02Integrate the site
03Operate the workflow
hardware + operations + human judgmentAn operating layer around the robot, the task, and the site.
Find your starting point 03 / Robots as a Service · RaaS
Robots as a Service. Beyond one vertical.
Hotels and data centers have different workflows. We start with yours: choose the right hardware, integrate it into your site, and support its operation as a service. Cloud kitchens and other verticals are the next environments we’re exploring.
Different sites.
A shared operating layer.
A basket in hand. A little help along the way.
A closer look
04 / The physics of helping
A little physics in every helping hand.
A useful robot needs ways to understand what’s around it, how it’s moving, and when to be gentle.
p = (x, y, z)A place in the world.
F = maForce, mass, and acceleration.
F = maForce, mass, and acceleration.
A little experiment for your curiosity. Choose a sense and move the slider.
Motion study / illustrativeFIG. 02
A little experiment in motion
Make every move count.
Joint positions and motion signals connect intent to a coordinated physical action.
p = f(θ)
Joint angles become a position in space.
YOUR TURN
Bend the elbow. One change at the joint draws a whole new position for the hand.
A simplified study of the idea. Move at your own pace.
05 / People in the loopYour hands.
Your hands.
A little more reach.
Human judgment, wherever the robot is. Our operator workspace brings approved people, assigned robots, and headset access together.
Open Live TeleopSign in with your approved operator account.
You + your headset
One operator session
Your assigned robot
FROM THE NOTEBOOK
Curious about what makes it work?
Cosmic 0.5 · Human-to-humanoid skill transfer, capture, retargeting, and evaluation.
Read the technical report06 / Physical AI models · evals & benchmarks
From model capability to field evidence.
Work with us to run evaluations and benchmarks on live deployment sites. See how your model behaves on a real robot, in a real environment, under the conditions that matter to the workflow.
Define the task, hardware, trial conditions, and operating limits together. Review outcomes, human interventions, and failure modes before expanding the deployment.
FIELD EVALUATION / PROTOCOL DESIGN
Measure what deployment demands.
- Task success
- Completion against agreed task criteria.
- Human interventions
- When and why an operator needs to step in.
- Cycle time
- Time to complete the physical workflow.
- Recovery behavior
- How the model responds when a task breaks down.
Example measures. Each benchmark protocol is scoped with the model team.
Read the technical report07 / Deployment & teleoperation datasets
Data from experience. For physical intelligence.
We collect real robot deployment data and have hundreds of thousands of hours of teleoperation data available for model teams.
Find data around the tasks, platforms, and intended use that matter to your research. Discuss coverage, sample availability, formats, and licensing with our team.
Experience from real operations.
Data collected from robots operating in real environments, with the context of actual tasks, sites, and human involvement.
- Source
- Real deployment sites
- Discuss
- Available tasks, platforms, and signals
- For model teams
- Model training, evaluation, and research
Request the inventory to discuss data fit and licensing.
Discuss your data requirements08 / For robotics hardware companies
You build the robot. We bring it to work.
Join a network of more than 15 robotics brand partners. We connect hardware with site operators, deployment operations, and teams developing physical AI models.
Bring your platform and its interfaces. We’ll discuss hardware fit, integration, human support, and a practical path into real operating environments.
ONE NETWORK / MULTIPLE BRANDS
Built to scale across brands and sites.
- 15+ robotics brand partners
- A deployment network built around hardware diversity.
- Real operating environments
- Hotels and data centers; cloud kitchens under exploration.
- Deployment + learning
- Operations, human support, evaluation, and data.
Our catalog is a reference guide. Listed platforms are not a count of partnerships or confirmed integrations.
09 / Out in the world
Part of a bigger conversation.
A few perspectives on the work, the people, and the questions moving robotics forward.
Visit the newsroom
Ban on Chinese robots leaves U.S. startups stranded
A look at the manufacturing challenges facing U.S. robotics startups, with perspective from CosmicBrain founder Anto.
By Viola ZhouRead the story
Opens on Rest of World in a new tab.
The 16 top logistics, manufacturing, materials startups from Disrupt Startup Battlefield
CosmicBrain in the Startup Battlefield selection.
By Julie BortRead the story
Opens on TechCrunch in a new tab.10 / Glad you asked
Curiosity looks good on you.
A few things you might be wondering.
01Can model companies run evaluations and benchmarks with you?
Yes. We work with physical AI model teams to run evaluations and benchmarks on live deployment sites. Hardware, tasks, conditions, and reporting criteria are scoped for each engagement.
02What robotics data do you offer?
We offer real deployment data and hundreds of thousands of hours of teleoperation data. Contact us to discuss available inventory, task coverage, formats, samples, and licensing.
03Are cloud kitchen deployments established today?
Cloud kitchens and a few other verticals are under exploration. We’re currently serving hospitality environments and deploying into data centers. New workflows begin with feasibility and pilot scoping.
04So, what does CosmicBrain actually do?
CosmicBrain is an operating system for deploying and operating robots across brands and sites. Our core business is Robots as a Service. We also work with physical AI model companies on live-site evaluations, benchmarks, and datasets.
05Do you build the robots, too?
We work with robot makers. Our focus is the software, data, integration, and human operations that help their hardware become useful in real settings. You can explore platforms and their source specifications in our robot catalog.
06Who do you work with?
Site operators including hotels and data centers; physical AI model companies needing evaluations, benchmarks, and data; and robot manufacturers looking for a deployment partner. Cloud kitchens and other verticals are under exploration.
07Are the robots fully autonomous?
Autonomy depends on the hardware, task, and environment. Human operators provide supervision and teleoperation where needed. We agree on the operating limits and evaluate a workflow before expanding it.
08How do we get started?
Tell us about the task you're working on, the setting, and what a useful result would look like. We'll talk through hardware, data, supervision, and a practical next step together.
11 / Your next chapter
What could we build together?
A robot deployment, a model evaluation, a dataset, or a hardware partnership. Tell us what you’re building and what a useful next step would look like.
human curiosity
× thoughtful engineering
a world of possibility
× thoughtful engineering
a world of possibility
People at the heart.
Robots in the loop.