Use AM-Bench¶

Choose a workflow by the result you need. Each page states its prerequisites, command, expected output, and validation boundary.

Run and inspect an environment¶

Run an Environment without a learned policy to check scene startup, observations, actions, cameras, and simulator performance.

Collect and validate data¶

Collect Demonstrations records scripted or teleoperated episodes in the canonical LeRobot format. Validate Datasets checks finalized metadata, episode boundaries, feature shapes, samples, action semantics, and logical resampling before conversion or training.

Evaluate a supported policy¶

Evaluate a Supported Policy covers the shared workflow and routes to ACT, Diffusion Policy, and OpenPI. Use Reproduce Research Workflows when preserving a complete experiment record.

Extend the system¶

Extend an Existing Environment covers task, robot, controller, and policy integrations. Start from the nearest maintained implementation and validate one layer at a time.