classify-blocks-2-1

A LeRobot dataset of 512 simulated single-block classification demonstrations. It preserves the original classify-blocks-2-512 recipes while revising the manipulation trajectory for hardware-facing geometry:

• approach_source holds at an 85 mm source hover with bounded lateral and height error;
• align_and_grasp descends through a 25 mm intermediate pose before table contact;
• align_target_slot stops 50 mm above the destination and lower_and_release does not descend further;
• retreat_and_reset closes the gripper and returns the arm to the episode's initial TCP position and full six-joint configuration within the recorded tolerance.

The original classify-blocks-2-512 dataset remains unchanged. This revision changes the motion contract, not the scene, class table, appearance randomization, or export schema.

Layout

The root follows the LeRobot dataset layout: meta/, data/, and videos/observation.images.front/. Additional review artifacts are:

• sidecars/<episode>/task_metadata.json: class materials, slots, legal target, and fact rules.
• sidecars/<episode>/frame_states.parquet: one row per exported frame.
• sidecars/<episode>/episode_receipt.json: immutable build receipt and file hashes.
• provenance/: plan, config, lineage, and aggregate validation.

The public aggregate deliberately omits the separate HUD preview videos. The LeRobot front camera video remains the authoritative policy observation.

Interface

The policy-facing observation is one 640x480 front image plus the six joint state values. The action is the absolute six-joint command at 15 Hz. training.action_valid is a supervision mask, not a policy input.

Scope

All 512 episodes are Cycles renders from the frozen classify-blocks scene. The trajectories are deterministic and the plant removes the measured random residual. This dataset tests scene variation, class matching, export, annotation, and training plumbing; it does not claim physical grasp reliability or hardware success.