Skeletal Animation: How Bones Move Game Characters

Almost every character you control in a 3D game moves with skeletal animation. An animator never moves the thousands of vertices in a character mesh by hand. Instead, the mesh is bound to a hidden skeleton of bones, the animator poses the bones, and the mesh follows. This guide explains how skeletal animation works, how it gets from Blender or Maya into Unreal Engine, where it breaks, and what to learn first if you want to work on it.

What is skeletal animation?

Skeletal animation (also called bone animation) is a way of animating a 3D model in two layers:

The mesh: the visible surface of the character, made of polygons, with its UVs and textures.
The skeleton: a hierarchy of bones (Maya calls them joints) that sits inside the mesh and is never rendered.

Each vertex of the mesh is linked to one or more bones. When a bone rotates, the vertices linked to it move with it. Animate the skeleton, and you have animated the character.

The alternative is to store the position of every vertex on every frame. That is called vertex animation, and it gets heavy fast: a 20,000-vertex character at 30 frames a second means 600,000 vertex positions for every second of motion. A skeleton with 60 or so bones stores a rotation (and sometimes a position and scale) per bone per frame instead. It is far lighter, and the same animation can play on any mesh that uses the same skeleton. That is why skeletal animation is the standard for game characters. The same idea is used in 2D games too, where flat image parts are bound to bones instead of a 3D mesh.

The three parts: skeleton, skinning, animation

A skeletal character is built in three steps. In a studio they are often done by different people.

1. The skeleton (rigging)

The rigger places bones inside the mesh and connects them in a parent and child hierarchy. The root usually sits at the floor between the feet, with the pelvis or hips above it. The spine, legs and arms branch out from there, and fingers branch from the hands. When a parent bone moves, all of its children move with it: rotate the upper arm and the forearm, hand and fingers follow.

On top of the skeleton, the rigger adds controls for the animator, such as inverse kinematics (IK) for legs, so the animator can plant a foot and move the hips without the foot sliding. If you want to build one yourself, our guide to rigging a character in Blender walks through it step by step.

2. Skinning (binding the mesh to the bones)

Skinning decides how much each bone moves each vertex. These amounts are called skin weights, and they run from 0 (no influence) to 1 (full influence). A vertex in the middle of the forearm might be 100 percent forearm. A vertex on the elbow might be split between the upper arm and the forearm, so it bends smoothly instead of creasing. In most setups the weights on each vertex are normalised so they add up to 1.

Most of what makes a character look good or bad in motion comes down to skinning. A clean skeleton with poor weights still gives you collapsing shoulders and knees that fold like paper. Our guide to Blender weight painting covers how to paint and fix weights for games.

Rigging and skinning are core technical artist work. In CADA's Certificate in 3D for Games, this is what the Tech Art specialisation covers, with rigging in Maya and Python for tools. The Animation specialisation, also in Maya, covers keyframing, timing and weight.

3. The animation

The animator poses the rig at key moments in time (keyframes) and the software fills in the frames between them. The animator then shapes the motion with timing, spacing and curves in the graph editor. The result is stored as rotation data per bone over time: a walk cycle, an idle, a jump, a sword swing.

Because the animation lives on the bones, not on the mesh, one walk cycle can drive any character built on the same skeleton. A game might have twenty different soldiers sharing one skeleton and one set of animations.

How skinning works under the hood

You do not need the maths to animate, but knowing roughly what the engine does explains most of the problems you will hit.

On every frame, the engine works out where each bone is, then moves each vertex by blending the transforms of the bones that influence it, using the skin weights. This method is called linear blend skinning, and it is what game engines use by default because it is fast and runs well on the graphics card.

It has one well-known weakness. When a joint twists a long way, such as a forearm rotating to turn the palm over, the blended vertices pull towards the middle and the limb thins out like a twisted sweet wrapper. Riggers call this the candy-wrapper effect. The usual fix in games is twist bones: extra bones along the forearm and upper arm that take part of the rotation, so no single joint has to twist too far. Dual quaternion skinning avoids the problem mathematically, and Blender and Maya both offer it, but it can bulge joints in other ways and it is not what most game engines use by default, so do not rely on it to fix a deformation you will see in-game.

Engines also limit how many bones can influence a single vertex. Four is the classic number and still a safe target for mobile and lower-end hardware. Before export, check your weights against the limit your project uses, and clean up tiny stray influences. If you let the engine drop extra influences for you, it will, and the result is often small jumps in the mesh that are hard to trace.

Skeletal animation in Unreal Engine

When you import a rigged character into Unreal Engine as an FBX file, you get a few separate assets. Knowing what each one does saves a lot of confusion.

Skeletal Mesh: the character mesh with its skin weights and materials.
Skeleton: the bone hierarchy. Several Skeletal Meshes can share one Skeleton, which is how they share animations.
Physics Asset: simple collision shapes on the bones, used for ragdolls and hit detection.
Animation Sequence: one clip of motion, such as a run or a reload, stored per bone.

The game then decides which animations to play and how to mix them, mostly in an Animation Blueprint. A few tools you will meet straight away:

Blend Spaces blend between clips based on values such as speed and direction, so a character moves smoothly from walk to jog to run.
State machines switch between states such as idle, moving, jumping and falling.
Montages play one-off actions, such as attacks or interactions, often on the upper body only while the legs keep running.
Retargeting copies animations from one skeleton to a different one with other proportions, so you do not have to animate everything twice.

One more term to know is root motion. Normally the game code moves the character and the animation plays on the spot. With root motion, the animation itself moves the root bone forwards, and the engine moves the character to match. It is common for attacks, rolls and climbing, where the exact distance of the move matters.

When games do not use skeletal animation

Skeletal animation is the default for characters and creatures, but not for everything. Good game artists know when to reach for something else:

Morph targets (shape keys in Blender, blend shapes in Maya) store whole sculpted shapes of the mesh and blend between them. They are common for facial expressions and corrective shapes on elbows and shoulders. Many game faces mix bones and morph targets.
Vertex animation baked into textures is used for big crowds, flags and foliage, where running a full skeleton on thousands of objects would cost too much.
Simulation handles cloth, hair and destruction, often driven on top of the skeletal motion.
Simple object animation moves a door, a lever or a turret as a whole, with no skinning at all.

Common beginner mistakes

Most skeletal animation problems are set up long before anyone animates. These are the most common:

Too few edge loops at the joints. A mesh can only bend where it has geometry. Elbows, knees, shoulders and hips need extra loops, or no amount of weight painting will save them.
Skinning in a pose that hides problems. Test every joint at its full range as you paint. A shoulder that looks fine at rest can fall apart when the arm goes above the head.
Unapplied scale and rotation. In Blender especially, a mesh or armature with unapplied transforms often imports into Unreal at the wrong size or with bones pointing the wrong way.
Too many bones. Every bone costs performance. Add twist and helper bones where the deformation needs them, not everywhere.
Bad bone naming. Clear, consistent names (with left and right suffixes) make mirroring, retargeting and teamwork far easier. Messy names cause pain all the way down the pipeline.

What to learn first

If skeletal animation interests you, work through it in this order:

Get comfortable in a 3D package. You need to move around, select and model before rigging makes sense.
Rig a simple character. A low poly figure with a basic skeleton teaches you hierarchy, pivots and IK without fighting a complex mesh.
Skin it and break it. Paint weights, pose every joint to its limits and fix what goes wrong. This is where most of your real learning happens.
Animate a walk cycle and an idle. Even if you want to be a rigger, animating your own rig shows you what animators need from it.
Get it into Unreal. Import the character, set up a Blend Space from idle to walk, and watch it move in a level. Seeing the whole chain work end to end is when it clicks.

Skip advanced facial rigs, muscle systems and custom auto-riggers for now. They are built on everything above, and they make much more sense once the basics are second nature.

Where to go from here

If you are new to 3D, start with the free Blender 3D Bootcamp. It is six live online sessions over three weeks with no experience needed, covering Blender basics and finishing with a session in Unreal Engine.

If you are ready to commit to game production, the Certificate in 3D for Games is a 60-week online programme where you try 3D Art, Tech Art, Animation and Procedural before choosing a specialisation. Each term ends in a team project: a walkable story environment, a playable level and a vertical slice. If you choose Tech Art, you build the rigs. No 3D experience or portfolio is needed to apply.