Skip to content
Robotics Glossary

Coordinate Frames

Practical

Coordinate frames (or reference frames) are the foundation for how robots understand position in space. A frame consists of an origin point and three orthogonal axes (x, y, z) that define directions of measurement. Every position, velocity, or orientation in robotics is defined relative to some frame.

What Is a Coordinate Frame?

A coordinate frame defines:

  • An origin — the (0, 0, 0) point where measurements start
  • Three orthogonal axes — x, y, z directions
  • A handedness — typically right-handed in robotics
         z (up)
         │
         │
         │
         ●───────► y (left)
        /origin
       /
      ↙
     x (forward)


  Right-Hand Rule (ROS Convention)

Frames can be fixed (attached to the world) or moving (attached to a robot or sensor).

Why Frames Matter

  • Sensors report locally: A camera sees objects in its frame — you must transform to the robot frame to act
  • Multiple perspectives: “1 meter to the left” depends on whose left
  • Kinematic chains: Robot arms have frames for each link: base → link1 → link2 → gripper
  • Sensor fusion: Without consistent frames, you can’t combine data from multiple sensors

Standard Frames in Robotics

map                    ← World-fixed (SLAM corrected)
 └── odom              ← Robot starting point (drifts over time)
      └── base_link    ← Robot body center
           ├── camera_link
           │    └── camera_optical_frame
           ├── lidar_link
           └── arm_base
                └── ... → gripper_link
FrameDescriptionROS Name
World/MapFixed global reference, does not movemap
OdometryTracks motion from start, continuous but driftsodom
Robot BaseAttached to robot body, moves with robotbase_link
SensorLocal to each sensor (camera, LiDAR, IMU)*_link
End-EffectorAt gripper or tool tipgripper_link

Frame Conventions

Different systems use different axis conventions:

Convention    X-axis     Y-axis     Z-axis     Used By
───────────────────────────────────────────────────────
FLU (ROS)     Forward    Left       Up         ROS, ground robots
FRD           Forward    Right      Down       PX4, aircraft
NED           North      East       Down       Aviation, GPS
ENU           East       North      Up         Geographic systems
Camera        Right      Down       Forward    Computer vision
USD/Graphics  Right      Up         -Forward   NVIDIA Isaac, Blender

TF2: The Frame Tree

ROS 2 uses TF2 to manage frames as a directed tree:

  • Each frame has exactly one parent (except the root)
  • No cycles — unambiguous path between any two frames
  • Time-buffered — query where a frame was at a past time

Static vs Dynamic

  • Static transforms (/tf_static): Fixed relationships like sensor mounts
  • Dynamic transforms (/tf): Changing relationships like odometry or joint positions

Working with Frames

Visualize the Frame Tree

Terminal window
# Generate PDF of current transform tree
ros2 run tf2_tools view_frames


# Print transform between two frames
ros2 run tf2_ros tf2_echo map base_link

Publish a Static Frame

Terminal window
ros2 run tf2_ros static_transform_publisher \
    --x 0.1 --y 0.0 --z 0.2 \
    --roll 0.0 --pitch 0.0 --yaw 0.0 \
    --frame-id base_link \
    --child-frame-id camera_link

Query Frames in Python

import rclpy
from rclpy.time import Time
from tf2_ros import Buffer, TransformListener


rclpy.init()
node = rclpy.create_node('frame_query')
tf_buffer = Buffer()
tf_listener = TransformListener(tf_buffer, node)


# Spin briefly to receive transforms
rclpy.spin_once(node, timeout_sec=1.0)


# Check if transform exists
can_transform = tf_buffer.can_transform('map', 'base_link', Time())
print(f"Can transform map→base_link: {can_transform}")


# List all known frames
print(tf_buffer.all_frames_as_yaml())

Common Mistakes

  1. Wrong frame assumption: Data is in a different frame than expected
  2. Forgetting camera conventions: Optical frames are oriented differently
  3. Ignoring time: Transforms change — use the correct timestamp
  4. Mixing conventions: ROS uses meters/radians, other systems may differ
Transforms Mathematical operations to convert between frames
Kinematics Computing positions through kinematic chains

Learn More

Sources