Why Humaan

Improving a robot starts with understanding what happened. Observations and actions show part of an attempt; its task, outcome, and human assistance explain how that attempt should inform the next learning cycle.

Humaan brings that context into a local workspace, where developers can inspect experiences, record judgments, and explain why particular examples belong in a learning set.

Preserve Experience and Context

A recording should remain useful beyond its first playback. Developers need to know which robot and task it describes, which files belong to it, and whether those files have changed.

Humaan recording manifests declare artifact paths, formats, byte sizes, and checksums. Human reviews live beside the recording, so reviewing an attempt does not rewrite its original artifacts. This preserves source data while allowing understanding of an experience to evolve.

Curate with a Learning Goal

Collecting more data does not explain what a robot needs to learn. A learning set needs an objective and a reason for including each experience. A successful attempt may demonstrate a useful behavior; a failure may expose a limitation worth evaluating.

Humaan keeps outcome, human assistance, and review state distinct. A reviewed experience can still have an unknown outcome. Including an experience in a learning set does not mean every recorded action should be imitated.

Keep Learning Decisions Traceable

Reviews can change as people inspect an attempt more closely. A saved learning set pins the selected review revision and recording manifest, preserving the evidence used at curation time. Later edits do not silently change that selection.

These local records support inspection and reproducibility. They are editable files, rather than a tamper-proof audit system.

Build Toward a Continuous Loop

Humaan's direction connects experience, data preparation, training, evaluation, and deployment. Each cycle should make it possible to understand what changed and assess whether a robot's capability improved.

The current alpha connects Builder, local simulation, recording review, and limited joint-feedback training. Physical execution and automatic feedback remain planned. Learn how these workflows fit together in Core concepts.