homogeneous transformation matrix
A homogeneous transformation matrix is a single 4x4 grid of numbers that captures everything about how one frame of reference relates to another: both how it is rotated and how it is shifted in space. A frame is just an agreed-upon set of axes and an origin, like 'the corner of the table' or 'the robot's wrist.' The transform bundles the rotation part (which way the second frame is twisted relative to the first) and the translation part (how far away its origin sits) into one tidy object. Feed a point's coordinates through this matrix and it pops out expressed in the other frame, as if you had carried the point from one viewpoint to another.
Packing rotation and translation together into one 4x4 is the clever trick the word 'homogeneous' refers to. On its own, rotation can be done with multiplication, but adding a shift normally needs a separate addition step; by tacking on one extra row and treating positions as having a phantom fourth coordinate of one, the matrix lets a single multiplication do both at once. The real payoff is chaining: if you know the transform from the table to the robot base, from the base to the arm, and from the arm to the gripper, you simply multiply the three matrices in order to get the transform from the table straight to the gripper. This is how a robot figures out where its hand is in the world by walking through every joint.
Mathematicians call the set of all these valid rigid-body transforms SE(3), the special Euclidean group, which is just a precise name for 'every way you can move a solid object around without bending or stretching it.' In software, whole trees of these transforms describe a robot's body and its surroundings, and systems like ROS keep them organized in a transform tree so any part of the robot can ask 'where is that, from my point of view' and get an answer.
To find where a screw spotted by the wrist camera sits relative to the table, the robot multiplies the table-to-base, base-to-wrist, and wrist-to-camera transforms, then applies the result to the screw's coordinates.
Chaining a few transforms answers 'where is this thing, in that frame' with one multiplication.
The 4x4 form is mostly a packaging convenience; the top-left 3x3 block is the rotation and the right-hand column is the translation, with a fixed bottom row of 0 0 0 1.