Unreal Motion Matching: Build Searchable Motion Before Tuning Weights

Understand the selection problem

Unreal's Motion Matching uses Pose Search to select animation poses from indexed animation data. A Pose Search Schema defines the features used for comparison, and a Pose Search Database supplies the animation assets. The Motion Matching node uses that data inside an Animation Blueprint. The system selects available motion; it does not automatically author an animation that the database lacks. [1]

Make the query and database describe the same movement

Schema channels can describe features such as bone positions, velocities, and trajectory samples. Their weights influence matching, while the selected features determine which differences the search can observe. [1]

Choose features that matter to the movement being built. For a grounded courier, inspect foot motion and the intended travel path before adding unrelated bones. Check that the trajectory, skeleton, coordinate choices, and authored movement speeds describe compatible behavior. A weight adjustment cannot repair every mismatch between the movement model and the animation collection.

Diagnose a concrete cornering failure

Imagine a courier whose database contains straight walking, straight running, and stops, but no suitable sharp cornering motion. The player's intended path turns 90 degrees around a shelving unit. Increasing trajectory weight may prefer a different available segment, but it cannot supply the missing authored turn. This is an illustrative content-coverage problem.

First inspect whether appropriate turns exist and are eligible for search. Then check whether the trajectory predicts the intended turn and whether the schema captures the relevant motion. Only after those checks should weights and blending be tuned. Record the asset and pose chosen at the corner, rather than judging only the final blended silhouette. [1][2]

Filter eligibility with gameplay meaning

The Game Animation Sample demonstrates selecting which databases are searched through a Chooser. Its walking setup can restrict selection to walking databases, preventing a running pose from winning when walking is required. The sample also uses a Pose History node to provide trajectory and cached pose information. [3]

Apply that separation deliberately: gameplay determines whether a movement family is allowed; matching chooses within the permitted material. A carrying or injured movement family needs a clear eligibility rule and suitable content. Do not expect a numeric preference alone to enforce every hard gameplay constraint.

Inspect decisions and budget the data

Use Motion Matching's Rewind Debugger integration to inspect recorded active, continuing, and candidate poses, together with channel breakdowns and trajectory drawings. Matching cost is a selection score derived from features; it is not a CPU duration measurement. [1][2]

Higher database sampling rates and additional channels can increase memory use and processing work. Add detail to resolve a demonstrated selection problem, then measure the resulting workload on the target. Review starts, stops, corners, and movement-family changes. A favorable score still needs visual review of transitions and contact quality. [1]

Sources

1. Motion Matching — Epic Games.
2. Motion Matching Debugging — Epic Games.
3. Game Animation Sample Project — Epic Games.