franka-demos-compressed

Franka Emika teleoperated demonstrations recorded as ROS 2 bags, with a synchronised HDF5 conversion.

RGB is stored as raw JPEG bytes (encoding='jpeg' attribute), which makes this split roughly 4x smaller than the decoded variant at fm-dev/franka-demos. Depth, state and timestamps are byte-for-byte the same in both.

Contents

raw/<task>/<operator>/<episode>/   # original ROS 2 bags (.db3.zstd) + metadata.yaml + task.txt
h5/<task>/<episode>.h5             # converted, time-synchronised HDF5

Task	Episodes	Prompt
button_order	52	press the buttons with the same order shown in the video
pick3	60	pick 3 times

H5 layout

One group per synchronised sample, frame_000000 ... frame_NNNNNN, each containing:

• Cameras (cam_base, cam_hand): colour image, aligned_depth_to_color as uint16 (720x1280), and the matching camera_info.
• Franka state: current_pose (position / orientation), measured_joint_states, desired_joint_states, external_joint_torques, external_wrench_in_base_frame, external_wrench_in_stiffness_frame, desired_end_effector_twist, last_desired_pose.
• Commands: _target_pose_position, _target_pose_orientation, _target_joint.
• Gripper / teleop: _franka_gripper_joint_states, _spacenav_joy_buttons.
• Timing: original_timestamps, camera_timestamps, other_timestamps, plus a sync_quality group with the per-topic offset from the reference RGB stream, a stale_topics list and a depth_stale flag.

Frames are synchronised onto the cam_base colour stream at full rate (no decimation). Non-reference topics are matched within +-100 ms and otherwise best-effort filled from the nearest message (capped at 1 s) with the frame flagged stale, so a good RGB frame is never dropped because one stream skipped a beat. Sample 0 of each episode is dropped.

Converted with rosbag_to_h5_simple.py.

import cv2, h5py, numpy as np
with h5py.File("h5/pick3/pick3_20260826_205000_123.h5", "r") as f:
    g = f["frame_000000"]
    rgb = cv2.imdecode(g["_cam_base_camera_color_image_raw_compressed"][:], cv2.IMREAD_COLOR)
    depth = g["_cam_base_camera_aligned_depth_to_color_image_raw_compressedDepth"][:]