Teach by demonstration

Show it once → teach the robot the task.

A person demonstrates how to perform a task. MANUSH converts the demonstration into a reusable robot skill. Traditional automation says program the robot; MANUSH aims for describe the task → demonstrate the task → robot learns the task.

In the developer dashboard

Open Skills → Teach by demonstration and describe:

1. The task — name, category, target robot and the demonstrated steps (one per line).
2. Objects and tools involved, the workspace and whether people work nearby.
3. Recording metadata — recording type (teleoperation, kinesthetic teaching, third-person video, egocentric video or motion capture), duration, FPS, frames, captured sensors and a reference to the recording.

What MANUSH generates

A draft skill with version 0.1.0 and a manifest containing:

• a policy outline where every step is mapped to a primitive — navigate, grasp, place, fasten, weld, actuate, inspect, manipulate, wait or speak;
• required sensors from the task category and what you captured;
• a conservative SafeCore envelope — slow speed (0.5 m/s or less when people are nearby), a category torque cap, a larger human distance, only the tools you named, human supervision and remote shutdown on;
• the MANUSH OS modules the skill will use;
• the demonstration metadata for reviewers.

Next steps

1. Review and tune the manifest in the version editor.
2. Evaluate the skill with synthetic runs in MANUSH Sim.
3. Train the policy with the MANUSH SDK and upload the package as a new version.
4. Submit for SafeCore review.