Position, Orientation, and Coordinate Frames

Opening challenge: Can the same physical point have two different—but equally correct—coordinates?

The Four-Part Pose Contract

A coordinate tuple is incomplete until you know its reference frame. For moving robots, time matters too. A decision-ready pose answers four questions:

1. Where? Position $(x,y,z)$.
2. Which way? Orientation, often represented by a quaternion.
3. Relative to what? A named coordinate frame.
4. When? The timestamp at which the relationship was valid.

ROS standard conventions use right-handed frames. For a body frame, the common convention is $x$ forward, $y$ left, and $z$ up. A point at (1, 0, 0) in base_link is therefore one metre ahead of the robot—not one metre east in the map.

Position Is Not Pose

Position locates a point. Orientation describes how axes are rotated. Together they form a pose. ROS geometry_msgs/Pose contains a Point position and Quaternion orientation; stamped variants add a frame identifier and time through a header.

Roll, pitch, and yaw are intuitive, but rotation order and axis convention must be stated. Euler-angle representations can become singular. Unit quaternions compose and interpolate rotations without that singularity, but their four components are not four angles, they must be normalized, and $q$ and $-q$ encode the same rotation.

Transform a Point: Rotate, Then Translate

Suppose the robot base is at map position $(2,1)$ with yaw $+90^\circ$. A target is one metre forward in base_link, so:

$p_{base} = (1,0)$

Rotate it by $90^\circ$:

$R p_{base} = (0,1)$

Then add the base translation:

$p_{map} = (2,1) + (0,1) = (2,2)$

The physical point did not move. Only its description changed. Reversing the operation requires the inverse transform; merely subtracting coordinates from different frames is invalid.

Frames Form a Time-Varying Tree

ROS tf2 tracks relationships among frames over time and transforms stamped data between them. A mobile robot commonly uses a chain such as:

map → odom → base_link → sensor or tool

REP 105 distinguishes a globally stable map frame from a locally continuous odom frame, while base_link stays rigidly attached to the robot body. Sensors and tools have their own calibrated frames. The transform requested must match the data timestamp; “latest” can be wrong for a moving platform.

Five Checks Before Combining Geometry

• Are source and target frames named?
• Are units and axis conventions compatible?
• Is the transform direction correct?
• Is the transform valid at the measurement time?
• Was rotation applied before translation, or was the correct homogeneous transform used?

Retrieval Pause

1. What four facts make a pose decision-ready?
2. Why can one point have two correct coordinate tuples?
3. A point is (1, 0) in base_link; the base is at map (2, 1) with yaw +90°. What is the map coordinate?

Sources and Further Reading

• Open Robotics, REP 103: Standard Units of Measure and Coordinate Conventions, retrieved 2026-08-18.
• Open Robotics, REP 105: Coordinate Frames for Mobile Platforms, retrieved 2026-08-18.
• ROS 2, geometry2 / tf2, retrieved 2026-08-18.
• ROS 2, common_interfaces / geometry_msgs, retrieved 2026-08-18.

These sources are linked and paraphrased; no third-party media is reproduced.