conversion_config.yaml

# =============================================================================
# OpenArm MCAP to LeRobot Conversion Configuration
# =============================================================================
# Teleop Mode: Leader-Follower
#
# In this mode, both leader (action) and follower (observation) joints
# are published on a single /joint_states topic. The converter uses
# joint name prefixes ("leader_"/"follower_") to separate them.
#
# For Quest teleop mode, see: openarm_bimanual_quest.yaml
# =============================================================================

# ROS2 Topic containing all joint states (sensor_msgs/JointState)
robot_state_topic: "/joint_states"


# =============================================================================
# Joint Name Parsing
# =============================================================================
# Your joint names look like: "leader_r_joint1", "follower_l_finger_joint1"
# Pattern: {source}_{arm}_{joint_id}
#
# Example parsing:
#   "leader_r_joint1"       -> action, right arm, joint1
#   "follower_l_finger_joint1" -> observation, left arm, finger_joint1

joint_names:
  separator: "_"

  # First part: data source -> determines observation vs action
  #   leader/master joints   = action (target positions to reach)
  #   follower/puppet joints = observation (current robot state)
  source:
    leader: action
    follower: observation

  # Second part: arm identifier -> for bimanual robots
  # Set to {} for single-arm robots
  arms:
    r: right
    l: left


# =============================================================================
# Camera Configuration
# =============================================================================

# ROS2 image topics (sensor_msgs/Image or sensor_msgs/CompressedImage)
camera_topics:
  - "/cam_waist/image_raw/compressed"
  - "/cam_wrist_r/image_raw/compressed"
  - "/cam_chest/image_raw/compressed"
  - "/cam_wrist_l/image_raw/compressed"

# Map ROS topics to dataset camera names
# These become: observation.images.{name}
camera_topic_mapping:
  "/cam_waist/image_raw/compressed": "waist"
  "/cam_wrist_r/image_raw/compressed": "wrist_r"
  "/cam_chest/image_raw/compressed": "chest"
  "/cam_wrist_l/image_raw/compressed": "wrist_l"

# Target image resolution [width, height]
image_resolution: [640, 480]


# =============================================================================
# Feature Extraction
# =============================================================================

# What to extract from observation joints (follower)
observation_feature_mapping:
  state: "position"      # Primary feature -> observation.state
  others:                # Additional features
    - "velocity"         # -> observation.velocity
    - "effort"           # -> observation.effort

# What to extract from action joints (leader)
action_feature_mapping:
  state: "position"      # Primary feature -> action
  others: []             # No additional action features


# =============================================================================
# Quick Reference
# =============================================================================
#
# Input (ROS JointState message):
#   name: ["leader_r_joint1", "leader_r_joint2", ..., "follower_l_joint1", ...]
#   position: [0.1, 0.2, ..., 0.3, ...]
#   velocity: [0.01, 0.02, ..., 0.03, ...]
#   effort: [1.0, 2.0, ..., 3.0, ...]
#
# Output (LeRobot dataset):
#   observation.state     [16]  = [left_joints..., right_joints...]  (from follower)
#   observation.velocity  [16]  = [left_joints..., right_joints...]  (from follower)
#   observation.effort    [16]  = [left_joints..., right_joints...]  (from follower)
#   action                [16]  = [left_joints..., right_joints...]  (from leader)
#   observation.images.waist    = camera0 frames
#   observation.images.chest    = camera2 frames
#   observation.images.wrist_r  = camera1 frames
#   observation.images.wrist_l  = camera3 frames