• 1. Franka Data Collection
  • 1.1. Overview
  • 1.2. Configuration
  • 1.3. Running
    • 1.3.1. Launch the Node
    • 1.3.2. Start/Stop a Recording
  • 1.4. Output Layout

1. Franka Data Collection

A ROS 2 package that records synchronized robot and camera data.

1.1. Overview

The data_collector node buffers synced RGB(-D) frames from a RigSnapshot topic in memory and, on trigger, simultaneously:

• records the configured ROS topics (joint states, poses, etc.) to an mcap rosbag via ros2 bag record, and
• encodes the buffered camera frames to per-camera video files via ffmpeg in a background thread, so the next episode can start immediately without waiting for encoding to finish.

1.2. Configuration

All settings live in config/data_collector.yaml:

storage_path: "/path/to/dataset"

topics:
  - "/franka_right/franka/joint_states"
  - "/franka_right/target_pose"
  # ...

videos:
  topic: "/zed_rig/synced_raw_snapshot"
  camera_fps: 30.0
  cameras:
    - name: zed_middle
      type: rgb
  codec:
    rgb: h264
    depth: ffv1
    crf: 23

• topics: plain ROS topics recorded straight into the rosbag.
• videos.topic: the franka_custom_msgs/RigSnapshot topic providing the synced camera frames to buffer and encode.

1.3. Running

1.3.1. Launch the Node

ros2 launch franka_data_collection start.launch.py

Note

Pass config_path:=/path/to/your.yaml to override the default config.

1.3.2. Start/Stop a Recording

Recording is toggled by calling the trigger service — the first call starts it, the second stops it:

ros2 service call /data_collector/record_data_trigger std_srvs/srv/Trigger

1.4. Output Layout

Each triggered episode is written under storage_path as:

episode_<timestamp>/
├── mcap/                     # rosbag with the configured topics
└── videos/
    ├── timestamps.npy        # int64 ns, one per synced frame
    ├── rgb_<camera_name>.mp4
    └── depth_<camera_name>.mkv   # only when depth cameras are configured