Does pose include speed?

No. Speed and angular velocity describe motion and require additional quantities.

Is pose always three-dimensional?

No. A planar model often suffices for x, y and heading, within its assumptions.

Is human pose the same representation?

Human-pose estimation often concerns multiple body keypoints or joints. This entry describes rigid-object pose, rather than reducing an articulated person to one rigid transform.

AI Glossary / Graphs and symbolic AI / Advanced

Pose

Definition

Pose is an object's position and orientation relative to a specified reference frame. It describes both where the object is and how it is turned. In robotics and computer vision, the object may be a robot body, camera, tool or item being handled. A pose alone does not describe its velocity, acceleration or internal joint arrangement.

What it means

Two delivery robots can occupy the same position while facing different directions. They have different poses, and the same command to move forward would take them toward different places. Similarly, placing a camera at the correct location is insufficient if it points away from the subject. Pose combines location and rotation into one spatial description. Its numbers require context: the coordinate frame, units and rotation convention determine what they mean. An estimated pose is also distinct from perfectly known ground truth.

How it works

A free rigid body in three-dimensional space has six degrees of freedom: three translational and three rotational. A planar pose commonly uses x, y and one heading angle. In three dimensions, orientation can be represented by a rotation matrix or unit quaternion, among other choices. Four quaternion components do not create four independent rotational degrees of freedom.
A rigid transformation combines rotation R and translation t. For a point expressed in an object frame, its coordinates in the reference frame are R times the point plus t. A homogeneous transformation matrix packages this operation so transformations can be composed. The direction matters: the transform from object coordinates to map coordinates differs from its inverse. Rotations and translations cannot generally be reordered without changing the result.

Three practical examples

A warehouse robot estimates its position and heading relative to a floor map. Its navigation system uses that pose when turning a route instruction into motion in the robot's local directions.
A robot arm needs the pose of a package relative to its working frame. The grasp depends on which face is upward and where the package sits, rather than on its center position alone.
An augmented-reality system estimates a camera pose so a virtual object stays aligned with the physical scene as the user moves. Errors in translation and orientation can produce different visible misalignments.

Think of it this way

An arrow pinned to a map conveys more than a dot. The dot supplies a location, while the arrow supplies a direction. In three dimensions, the object can also tilt and roll, so a single heading arrow no longer captures its complete orientation.

Worked example

Use a flat map with x pointing east and y pointing north, measured in meters. A robot's local x axis points forward and its local y axis points left. Its position in the map is (4, 2), and its heading is 90 degrees counterclockwise from map east. Its forward direction therefore points north.
A marker has robot-frame coordinates (2, 1): two meters forward and one meter left. At this heading, the rotation maps a local point (a, b) to (-b, a). Rotating the marker gives (-1, 2). Adding the robot's map position gives (-1, 2) + (4, 2) = (3, 4). That is the marker's map position.
If a program incorrectly adds coordinates without rotating, it obtains (6, 3). The discrepancy is (-3, 1), whose length is square-root(10), approximately 3.16 meters. A seemingly small convention error has changed the interpreted location substantially.
To invert the correct transformation, subtract the translation first: (3, 4) - (4, 2) = (-1, 2). Then rotate by minus 90 degrees, mapping (u, v) to (v, -u). This recovers (2, 1).
Now consider a small camera whose origin is (0.5, 0) in robot coordinates and whose heading is minus 90 degrees relative to the robot. Its map position is (4, 2.5), and its map heading is 0 degrees. It sits north of the robot's origin while pointing east. Position and orientation both require the frame relationships to be applied.

Illustrative case study

In a fictional fulfillment center, a dashboard shows a robot at the correct map location, but the displayed position of its detected marker is wrong. The team checks the map above and finds that one component treated robot-frame coordinates as map-frame offsets.
They create a regression example using the known pose and marker coordinates. The correct result must be (3, 4), and applying the inverse transform must recover (2, 1). They also test the camera's offset and orientation separately. These checks establish consistent coordinate handling for the example; further tests are required for sensor accuracy, calibration and motion during measurement.

Common misconceptions and mistakes

Position alone is not pose, and heading alone is not a complete three-dimensional orientation. A robot arm's tool pose does not necessarily identify every joint angle: different arm configurations can place the tool similarly.
Another mistake is mixing conventions. ROS's common body-frame convention uses x forward, y left and z up, while its camera optical convention uses z forward, x right and y down. The relationship must be represented explicitly. Degrees and radians must also be distinguished. Finally, a precisely formatted pose record can still contain an uncertain or outdated estimate.

Business use

Pose connects perception to physical action in navigation, inspection, manipulation and spatial interfaces. Define which object frame is being located, which reference frame is used and when the estimate applies. Document units and orientation representation at system boundaries. Measure translation error and orientation error appropriately instead of treating every stored number as interchangeable. A position error of centimeters and an angular error of degrees describe different quantities. For a long tool, an orientation error may shift the tool tip even when the reference point is correctly positioned.

Check your understanding

Where is the example marker in map coordinates?

Answer: At (3, 4) meters.

Why is simply adding its coordinates wrong?

Answer: The robot and map axes have different orientations.

Does the camera's eastward heading mean its origin lies east of the robot?

Answer: No; its origin lies north, at (4, 2.5).

Related terms

Orientation
Localization
Reference frame
rigid transformation
rotation matrix
quaternion
camera extrinsics
forward kinematics
six degrees of freedom

Learn next

Repeat the marker calculation with a zero-degree robot heading. Then compare transforming a point, which includes translation, with transforming a direction vector, which does not.

Sources

Lynch and Park, Modern Robotics, homogeneous transformation matrices: https://modernrobotics.northwestern.edu/nu-gm-book-resource/3-3-1-homogeneous-transformation-matrices/ ROS REP 103, coordinate conventions, official source: https://github.com/ros-infrastructure/rep/blob/master/rep-0103.rst AIMA glossary, pose: https://github.com/aimacode/aima-glossary/blob/master/glossary.md

Previous: Population variance Next: Positional embeddings