Why We’re Building Humaan

Our mission, the future we want to build, and the principles that will guide us.

We believe intelligence should help people in the physical world: in the places we work, the spaces we live, and the tasks that demand our time, effort, and care. Robots could extend what people are able to do, take on dangerous work, and help us respond to needs that are difficult to meet today.

Reaching that future means building machines that can learn from what happens around them. Machines that can improve as we understand their successes, investigate their failures, and give them better examples to learn from.

That is the future we want to help build with Humaan.

Reliability is built in the real world

A robot’s usefulness depends on how reliably it works under real operating conditions. Repeated failures and unplanned interruptions make it harder for people to depend on robots in everyday work. Closing that gap requires learning from the conditions in which the robot is expected to work.

Real-world deployment data is essential to that progress. It captures rare situations, failure modes, and recovery behaviors specific to the task and environment: where a robot’s learned behavior breaks down, when a person needs to intervene, and what it takes to complete the task after something goes wrong. These experiences are difficult to anticipate in full or obtain from internet data alone. They reveal the gaps that matter in operation.

That makes a continuous data loop central to our vision: deploy within validated limits, capture failures and interventions, curate useful examples, retrain, and evaluate before redeploying. As reliability improves, robots can be deployed more broadly, exposing them to a wider range of conditions. Those experiences can feed the next round of improvement. The flywheel turns when deployment experience becomes validated progress.

Data quality determines whether that loop produces useful learning. More recordings have limited value if they repeat familiar successes while missing consequential failures. The priority is to capture relevant situations with accurate observations, actions, and outcomes—including examples of successful recovery. Video, human demonstrations, simulation, and deployment data each contribute different kinds of evidence. Their value depends on how they are selected and combined to address the capabilities a robot still lacks.

Without a path from field experience back to training and evaluation, a failure may be fixed once on-site yet remain unresolved in the model. We are building Humaan to help make that path a foundation developers can build on.

Our mission

Humaan’s mission is to advance physical intelligence by enabling robots to learn from real-world experience.

We are building toward a learning layer for Physical AI: a foundation that helps developers connect robot experience with the work of improving robot intelligence.

For us, experience means what a robot observes, the actions it takes, and what happens as a result. Learning means making those experiences useful through deliberate selection, training, and evaluation. It requires human judgment, clear evidence, and repeated testing.

We want Humaan to help developers understand what a robot did, identify what it needs to learn, and assess whether a change improved its behavior.

Human dignity at the center

We believe every person has inherent worth. Physical intelligence should serve that worth and expand people’s ability to shape their own lives.

For Humaan, this means pursuing technology that reduces dangerous and exhausting work, supports human judgment, and gives people more time and opportunity for what matters to them. We will judge its value by how it improves people’s lives as well as what it enables machines to do.

People should have a voice in how robots enter their workplaces and communities. Their knowledge should inform development, their privacy should be respected when experience is collected, and they should retain meaningful oversight of systems that affect them. As work changes, the people affected deserve a voice in that transition and opportunities to learn and participate.

These principles will guide our choices about what we build, how we evaluate it, and whom it serves. Human dignity, safety, and agency belong at the center of progress in physical intelligence.

Make physical intelligence composable

Our long-term vision is to make physical intelligence composable.

We want developers to connect data, tools, and robot capabilities through clear interfaces, reuse useful work, and improve individual components without rebuilding the surrounding system each time.

Composability also means making the conditions of reuse understandable. A dataset or capability developed for one task will not necessarily work for another robot or environment. The assumptions, context, and evaluation behind it need to remain visible so that others can decide how to adapt it.

If we can make those connections easier, a small team can build on foundations it could not create alone. A researcher can make an experiment easier for others to inspect and extend. Work on one task can become a useful starting point for the next.

We want more people to contribute to physical intelligence, with progress that others can understand and carry forward.

Our commitment

We commit to making real-world experience a foundation for progress in physical intelligence. The decisions we make at Humaan will serve a shared purpose: helping people understand what robots encounter, turn that understanding into learning, and carry proven improvements into the next deployment.

We will measure progress by the capabilities people can rely on in practice. That means preserving the context behind the data, making failures visible, and demanding evidence that a change improves behavior under the conditions that matter.

We will let the needs of developers and the people affected by robots guide our priorities. We will communicate limitations honestly and use feedback from real work to challenge our assumptions.

Our commitment is to a future where experience becomes shared progress, and physical intelligence expands what people can do.

Building in the open

We are building Humaan as open-source software because developers should be able to inspect the tools they rely on, adapt them to their needs, and contribute improvements that others can use.

We believe developers should retain control over their data and the systems they build. Open interfaces and modular foundations should give them the freedom to choose how they work and carry their work forward.

We also want to build a community where the difficulties are visible. Failed attempts, unexpected behavior, and awkward workflows can help us understand what needs to change. Sharing those lessons is part of how we hope to make the tools more useful.

An invitation to build

The future of physical intelligence should be shaped by both the people who build it and the people who live and work with it.

Whether you develop robots, study how they learn, or understand the work they could help with, we invite you to build with us. Share what you learn, challenge our assumptions, and help create foundations that others can carry forward.

Explore Humaan on GitHub.

Let’s assemble what’s next, together.