← All Guides

Combat FIT · Learning Hub

AI Poomsae and Tul Analysis — How AI Pattern Coaching Works for Taekwondo

Poomsae and tul are the technical backbone of Taekwondo. Every grading, every competition, every serious practice session involves patterns. And yet for most students — even dedicated ones — the feedback loop is broken. You practice at home, you perform in class, your instructor corrects you, you practice at home again. What you almost never have is the ability to see yourself objectively, in real time, between classes.

That's the gap AI pattern coaching is designed to close.

The Core Problem: You Can't See Yourself

I've trained in both WTF (now World Taekwondo) and ITF systems. The problem is the same in both: self-correction without visual feedback is almost impossible. You might feel like your back stance is deep enough. You might believe your knife-hand block is chambering correctly. You might think your kihap is landing on the right technique. Without a mirror, a video, or a coach standing beside you, there's no way to know.

Mirrors help but only tell you what you look like in a static position — not whether your timing is correct, whether your centre of mass is where it needs to be through the transition, or whether you're hitting the right joint angles at peak extension. Video helps but only if you know what to look for. Most students don't have the analytical framework to critique their own footage at the level of technique their instructor expects.

How AI Pose Extraction Works

Modern AI pattern analysis uses pose estimation — specifically frameworks like MediaPipe — to extract a full skeletal model from video in real time. The system identifies key landmarks: shoulders, elbows, wrists, hips, knees, ankles, and the relationships between them. From these landmarks it calculates joint angles at every point in the movement — the angle of your front knee in a walking stance, the elevation of your elbow during a high block, the rotation of your hip through a turning kick.

It also captures movement timing: how long each stance transition takes, where the pause falls within a pattern, whether your speed-power techniques are actually accelerating through the strike. These are dimensions of performance that even experienced students struggle to assess from feel alone. The output is a structured data representation of your movement — one that can be compared directly against a reference model of correct technique.

WTF vs ITF: Different Systems, Different Standards

This distinction matters enormously and most AI tools ignore it entirely. WTF poomsae (the Taeguek series through to the higher Yudanja forms) and ITF tul (Chon-Ji through to Tong-Il) are different pattern systems with different stances, different techniques, and critically, different scoring criteria. A back stance in ITF is not the same as a back stance in WTF. The sine wave motion fundamental to ITF is entirely absent from WTF poomsae. A single AI model trained on WTF movement will give incorrect feedback to an ITF practitioner — and vice versa.

Any AI poomsae or tul coach worth using must maintain separate reference modelsfor each system and score conformance against the appropriate standard. This is not a minor technical detail — it's the difference between the tool being useful and actively harmful to a student's development.

What Conformance Scoring Actually Means

Conformance scoringis not a "looks good" assessment. It's a quantitative comparison of your movement against a reference execution of the same pattern — frame by frame, joint angle by joint angle, timing beat by timing beat. The score tells you specifically where you deviate from the reference and by how much.

This is different from aesthetic judging. A human judge at a poomsae competition is scoring their impression of your pattern against a mental model built from years of experience. A conformance score is a precise geometric comparison. The AI score is most useful as a training tool: it tells you where to direct your practice, not whether you'll place first on the day.

Practical Use: Between Classes and Before Gradings

The real value of AI pattern coaching is in the unsupervised training hours. You practice at home four times between classes. Without feedback, you may be ingraining errors. With AI analysis, you get objective feedback after every repetition — which specific techniques are drifting, which stances are collapsing under fatigue, where your timing is rushing.

In the two to three weeks before a grading, this becomes particularly valuable. You can isolate the sections of your pattern where your conformance score is lowest, drill those segments specifically, and track whether targeted practice is actually moving the score. That's a feedback loop that simply doesn't exist in traditional solo training.

Combat FIT's AI Pattern Coach

Combat FIT's AI Pattern Coach supports both WTF poomsae and ITF tul with system-specific reference models and conformance scoring — not a single generic framework applied to both. It also covers five karate kata styles (Shotokan, Kyokushin, Goju-Ryu, Wado-Ryu, and Shito-Ryu), recognising that technical requirements differ between styles. The coach is designed for home practice, integrating directly with your training log so your pattern scores sit alongside your mat time, grading history, and session notes.

Create a free Combat FIT account to access the AI Pattern Coach — available on PRO.

Written by Dr Sean Hall

MD, MBA, PhD — founder of Combat FIT. 30+ years on the mat across Taekwondo (WTF & ITF), BJJ, Kendo, and Iaido.

About Dr Sean Hall →

Apply this in Combat FIT

Start free with 30 days of PRO. Every new account gets 30 days of PRO. No card needed.

Start Free →