Techniques · Strikes
The hooks
The lead and rear hooks, the lead one from my own motion capture.
Where it stands
| Technique | Teacher kept | In the student | Own runs |
|---|---|---|---|
| lead hook | guard14-45.9M23 October 2026 | distill7-245.8M26 October 2026 | 3 |
| rear hook | training or prepared | distill7-245.8M26 October 2026 | 0 |
Both hooks started as references scripted from my spec in the five punches work: the elbow bent about 90° at shoulder height and locked to the chest, a small wind-up the other way, then the body’s turn carries the fist through. They hit within about 10 M steps, 6 of 6 each in the air at 5–6 m/s, but light on the heavy bag: 11–15 N s against a bar of about 21. The speed work brought the left hook to the target in 0.33 s instead of 0.52, though no harder.
The lead hook’s real story is the higher guard. On the new guard it fell to 0 of 12, and three times the presses didn’t bring it back. I said the left hook was wrong and asked for it to be rebuilt from my motion capture, filmed with two phones (Experiment01, shadowbox_01, 23.73–24.63 s). My hook isn’t the scripted one: the fist swings out wide with the arm nearly straight, then the hips, chest and closing elbow whip it in. My hips turn 27°, where the script asked for 70. guard5 already threw it that way and landed it 10 of 12 with no retraining.
Then I watched guard5’s hook closely: too low, the elbow dipping. It still landed, because the gate only counts hits. Rewards for the form left the swing about 25 cm low, the servos lagging on a swing of about 20 rad/s. guard11 added the reference’s feed-forward torque on the hook arm, guard12 let guard5 teach everything but the lead hook, and guard14 tripled the swing term. guard14-45.9M was kept: the elbow level with the fist, 12 of 12 in the air. It became the parent of the first body-straight and knee teachers, and guard18 after it is the base whose hooks every student has.
The rear hook is still the scripted one. Open: both hooks at the bag, and re-solving the wind-up, which lifts the fist above the head before it swings round and is the likely reason they’re slow.
What I think, as a coach
My spec put it in one line: “The power comes from the legs, hips and core, not the arm.”
On guard5’s lead hook: “too low. The elbow is kind of dipping, but the fist itself is also low. The hook should have the elbow the same level as the fist, and it should be hitting the opponent’s chin or the cheek.” At the impact it should be “coming completely from the side, 90 degrees”, not forward or at 45° like a jab. The elbow’s angle I leave free.
Every attempt
Every run that trained these techniques on its own, in order.
| Run | Outcome | Machines | Hours | Cost |
|---|---|---|---|---|
guard12-2026102423 October · lead hook · kept guard12-39.3M · The higher guard: retraining every skill onto a new stance | Kept | 5 | 1.5 | $1.34 |
guard13-2026102423 October · lead hook · kept guard13-29.5M · The higher guard: retraining every skill onto a new stance | Kept | 1 | 1 | $1.05 |
guard14-2026102423 October · lead hook · kept guard14-45.9M · The higher guard: retraining every skill onto a new stance | Kept | 4 | 1.8 | $1.65 |
| 3 training runs | 10 | 4.3 | $4.04 |
Shared training
Runs that trained it together with other techniques: a base model, a retrain onto a new stance, a distillation into the student. Their whole cost is counted once, in the shared figure above, not split.
How the models descend
Its own runs, each with an arrow from the run it started from; faded boxes are models from elsewhere it started from or learnt from. Tap a run for its details. The whole history is on the lineage page.