Skip to main content

Python packages overview

The se3labs SDK connects robot policies to SE3 stations. The SDK runs beside the model and exchanges observations and joint actions with the station over a session. The station controls the episode lifecycle, while its operator prepares the scene and supplies scoring answers.

The Python packages have two roles:

PackageImport namespacePurpose
se3labsse3labs.sdkClients and commands for running evaluations
se3labs-interfacese3labs.interfaceRobot policy contracts and shared data types

This section documents both packages' types, interfaces, and behavior. For setup and guided examples, use Tutorials.

Componentsโ€‹

ComponentResponsibility
JobRequestTask, robot policy, and requested episode count
PolicyInitialization, inference, and per-episode reset
Task_SpecTask instructions, rollout duration, and scoring questions
Evaluation clientSession transport and station status queries
SessionReportSession counters, close reason, and recorded policy errors

Referenceโ€‹

Release scopeโ€‹

v0.0.1a1 updates the SDK's interface dependency constraint to allow alpha releases. Its SDK implementation is unchanged from v0.0.1a0.

This release provides se3labs eval run, se3labs eval status, and the Python evaluation client. It predates browser authentication, CLI JSON result export, EndEpisode, the act harness, and XPolicyLab integration. Its SessionReport contains counters and a close reason, without terminal state or episode scores.

Use it with the corresponding station and relay release.