Inspiration

Treadmills are the most common excercise equipment for home use also are present in practically every gym. People walk for forty minutes, staring at a wall or a phone. I wanted to know what happens if the treadmill becomes the controller and the room becomes the world. With mixed reality this is finally made possible.

The idea that kept me going: what if every kilometer you walked actually took you somewhere? Not a step counter, but a journey, with villages, a quest-giver who needs your help, and a beacon at the end of every run.

What it does

Treadmill Warrior is a roguelite AR adventure for treadmills, with hand tracking and running in passthrough.

  • Step on any treadmill. There is no smart treadmill, no pairing and no wearables. The headset reads your stride from the small bob of your head.
  • A light trail unrolls through your room and carries you through worlds at the speed of your own pace.
  • Play with your hands. Catch wisps of light to fill your lantern, push a palm through shards of gloom, raise a hand to choose your road at forks, and press your palms together to light the beacon at the end.
  • Pip the firefly guides you, steers wisps into reach and dances when the beacon comes into view.
  • A real journey. Every completed run banks its distance toward the next village, 10 km away.
  • XP and levels reward distance, strides, wisps, gloom broken and beacons reached, with a bonus for running.

How I built it

The game is written in TypeScript on Meta's Immersive Web SDK (three.js and WebXR), built with Vite and served to the Quest browser. Everything you see is generated in code: no external 3D files are loaded at runtime.

Stride detection from the headset alone. The core problem was speed without treadmill data. When you walk on a belt you stay in place, but your head rises and falls once per footfall. I band-pass the headset's height to isolate that bounce, count a footfall each time it crosses a threshold, and only trust the rhythm after three regular strides. Cadence comes from the mean of the last six intervals:

$$ c = \frac{60}{\bar{t}}, \qquad \bar{t} = \frac{1}{6}\sum_{i=1}^{6} t_i $$

where $t_i$ are the gaps between footfalls in seconds, accepted between 0.22 s and 1.3 s.

Cadence becomes trail speed. A 120 strides-per-minute walk travels at 3 m/s, scaled within limits:

$$ v = 3 \cdot \operatorname{clamp}!\left(\frac{c}{120},\ 0.5,\ 1.6\right)\ \text{m/s} $$

A steady pace instead of a twitchy one. Following cadence directly made the world slow down every time I looked around. So the trail now holds a baseline pace learned in a warm-up, and only adopts a new pace when it stays outside a ±10% band long enough: 4 s to speed up, 8 s to slow down, because a false slow-down feels much worse than a false speed-up.

Progression.

XP per run is

$$ \text{XP} = \sum_{\text{events}} b(v)\cdot x_e, \qquad b(v) = \begin{cases} 1 & v < 2.5 \ 1.25 & 2.5 \le v < 3.5 \ 1.5 & v \ge 3.5 \end{cases} $$

with 100 XP per km, 1 per stride, 5 per wisp and 50 per gloom or beacon. Abandoned runs keep half. Level $L$ costs $1000 \cdot 1.25^{\,L-1}$ XP.

Worlds Scenery is instanced, low-poly and procedural: 48 slots recycle along a 72 m window, and each slot re-rolls its prop, position, size and turn every lap, seeded so it changes out of sight. Some worlds are glowing holograms; others (village cabins, crystals, dandelions) are flat-shaded with lighting baked into vertex colors, so dozens of copies stay cheap on a standalone headset.

Interface Menus use hold-to-press medallion buttons and a touch tray, so nothing needs a precise pinch mid-run. Story and dialogue text types out letter by letter, with recorded voiceover. Sound effects are synthesized live with the Web Audio API.

Testing I built a desktop preview that simulates strides, hands and gestures from the keyboard, and ran automated checks in a headless browser: playing whole runs, capturing screenshots and reading the game's state. That let me iterate many times a day and save headset sessions for what only a headset can tell you. In developmend I did use an AI assistant (Claude) extensively.

Challenges I ran into

  • Reading speed without sensors. A head bob is a few millimeters. Separating footfalls from nods, glances and turning to look at something took a lot of tuning: amplitude thresholds, interval limits and requiring a regular rhythm before trusting it.
  • Looking around would cause issues This led to the baseline-pace system described above.
  • Comfort and safety. The trail has to stay centered on a real treadmill with handrails. Scenery never comes within 2.1 m of the center line, the world can be recentered, and stopping eases the trail to a halt instead of freezing it.
  • Performance on a standalone headset. Every prop is instanced and every glow is baked rather than lit, so whole worlds render in a handful of draw calls.
  • Art direction. Not every model works on a moving trail. Some versions looked wrong in motion, and I rolled them back; the crystal horns, for example, only worked once they became a gateway arching over the trail.
  • Pacing a story around exercise. Distance only counts for completed runs, so a quest needs to fit several real workouts, not one session. The Dead Beacon fills over two to three runs.

Accomplishments that I'm proud of

  • It works on any treadmill with nothing but the headset.
  • Running feels like traveling: villages, worlds, quests.
  • Pip the Firefly, who became player's sidekick and heart of the game.

What I learned

  • The body is a great controller, if you design around it. Small, forgiving gestures such as a raised hand or palms together work mid-stride; precise ones don't.
  • Smoothing is a design decision Players notice a false slow-down far more than a slightly late speed-up, and the math should reflect that asymmetry.
  • Mixed reality is the right medium for fitness. Seeing your room keeps you safe, while the fantasy still lands.

I’ll answer the cadence question from the code rather than from memory. Checking that and fixing the canyon test together:

No gateway in this stretch, so a broader sample: five points across the 380 m canyon.

The arching horn pair now forms a gateway over the trail, with one horn on each side. I checked it in the desktop preview with no errors, and the 23 tests pass.

Horn gateway

Shape: each horn stands 3.2 m from the center of the trail, so they’re about 6.4 m apart, well clear of the treadmill. They’re slightly longer than on your sheet (4.6 m) and bend more, so the tips arch over your head about 4 m up, with a gap between them. Space around it: the spots on both sides near a gateway stay empty. In my checks, no other prop was within 4.5 m of one. How often: about one gateway per 72 m of canyon, never closer than 12 m apart. Some stretches have none. My first version allowed them 6 m apart, and a row of them looked like a tunnel. The other three props: the monolith, the cluster and the original crystal are unchanged and appear more often.

Zip: treadlight_horn_gate.zip

Your cadence question. There are no fixed speed steps: the trail speed follows your stride rate smoothly.

Speed range: at 120 strides per minute the trail moves at 3 m/s. It scales in proportion between half that (1.5 m/s, at 60 strides a minute or slower) and 1.6× (4.8 m/s, at 192 strides a minute or faster). Delay when you change pace: I simulated the game’s settings for a sudden belt change. Speeding up: walking at 100 to running at 160 strides a minute, the trail reaches its new speed in about 4.5 seconds. Slowing down: the reverse takes about 5.4 seconds. Slower strides arrive further apart, so the average catches up more slowly. Why the delay: your cadence is the average of your last 6 strides, and the trail then eases toward the speed that cadence calls for. If you stop stepping: the trail holds its speed for 2.5 seconds, so a missed step or two goes unnoticed. Then it eases to a stop, and the pause screen appears about 4.5 seconds after your last step. XP is the exception: the speed bonus does use three steps. Walking (under 2.5 m/s) gets ×1, jogging gets ×1.25, and running (3.5 m/s and above) gets ×1.5. Treadlight horn gate ZIP

To avoid constant slowing down due to looking around, it is best to not scale speed smoothy and instead use baseline speed as guidance. If the baseline speed and the new speed are in constant divergence then, after fixed amount of time game can smoothly move to new cadence and keep it as the new baseline. In general slowing down is more jarring, when in reality player is keeping same pace (than speeding up)

So the trail should hold a steady speed set by your baseline pace, and only change when your pace has clearly and consistently moved away from it, with slowing down needing more proof than speeding up. First, how the warm-up baseline is used now:

All 26 tests pass, including the three new ones. Now the same in the running game: start at the default pace, switch to running, then back to walking, and log the trail speed each second.

Now the trail settles on your actual pace: 177 when running, 105 when walking. Updating the README and packaging:

The trail now holds a steady baseline speed and only changes when your pace has clearly changed. It no longer follows every stride.

Baseline: the warm-up learns your pace, and the trail runs at that speed. Small wobbles are ignored: anything within ±10% of the baseline doesn’t change the speed, so glances, head turns and uneven strides do nothing. Speeding up: if your pace stays more than 10% faster for 4 seconds in a row, it becomes the new baseline. The trail then eases up to it over about 3 seconds. Slowing down: it needs 8 seconds of being more than 10% slower, then eases down more gently. Interruptions: dropping back inside the band, or losing the rhythm for a moment, restarts the count. Stopping still works as before: the speed holds for 2.5 seconds, then the trail eases to a stop. What the new baseline is: it’s your average pace over the second half of the wait, after your stride has settled. Without that, a slow-down left the trail slightly too fast (113 instead of 104).

In the desktop preview, switching to a run sped the trail up about 5 seconds later, and switching to a walk slowed it about 9 seconds later. The new baselines were 177 and 105, matching the pace I’d switched to. There were no errors, and all 26 tests pass, including 3 new ones: glances and dips leave the pace alone, the 4 s and 8 s waits, and an interruption restarting the count.

How it feels on the treadmill is the real test. The ±10% band and the 4 s and 8 s waits are each a single number in cadence.ts, so they’re easy to change. Changing your belt speed should feel deliberate. If a real change takes too long to register, shorten the waits. If looking around still nudges the speed, widen the band.

Zip: treadlight_steady_pace.zip

Treadlight steady pace ZIP

Let's add a finish scenario. Once player reaches Rybol, A message comes up saying. Congratulations you have finished the demo version of Threadmill Warrior. You can continue to play and level up or reset progress and start again."

A finish screen for reaching Rybol. First, how arriving at a village works at the end of a run, and how the summary’s buttons behave:

Create an elevator pitch for the game

Checks and types pass. Now the whole flow in the preview: a run that reaches Rybol, then the summary, then Next, then the message and both buttons.

The button logic itself is fine: after firing, a button waits until the hand moves away before it can fire again. So something else sent the game to the lobby. Tracing the phase changes:

Treadmill Warrior — in-game text Claude Sheets ·Only you

The end-of-demo screen is in. I checked the whole flow in the desktop preview with no errors, and the 26 tests pass.

When it appears: after the run that reaches Rybol, the summary shows your XP as usual, but its only button is Next. That opens the message panel:

Heading: “Rybol Village” Title: “Congratulations!” Message: “You have finished the demo version of Treadmill Warrior. You can continue to play and level up, or reset your progress and start again.”

A chime plays and Pip does its happy dance.

The buttons appear after 2 seconds. Without that wait, the hold that pressed Next could also pick an answer on the new screen; my first test did exactly that.

Keep Playing: back to the lobby, with your journey and XP kept. Start Again: the first hold turns it into Confirm?. A second hold within 5 seconds resets your whole profile, the same as Reset in Settings. It turns back if you don’t confirm.

I wrote “Treadmill” where your message said “Threadmill”. The spreadsheet has the new texts, and the README is updated.

Zip: treadlight_demo_complete.zip

Elevator pitch

One line: Treadmill Warrior turns your treadmill into an adventure: put on a Quest, start walking, and a glowing trail unrolls through your room, carrying you through worlds at the pace of your own stride.

About 30 seconds, spoken: Treadmill workouts are the most boring hour of the day. Treadmill Warrior makes them the best. It’s a mixed reality game for Meta Quest: you step on any treadmill, and a light trail appears in your room and carries you through glowing worlds. There’s nothing to pair and nothing to wear. The headset feels your stride from the bob of your head, so walk faster and the world moves faster. Along the way you catch wisps of light with your hands, push through gloom, raise a hand to choose your road at forks, and relight beacons, with Pip the firefly at your side. Every kilometer you walk counts toward a real journey between villages, with quests, levels and a story. It’s a workout that feels like a quest, not a chore.

Why it stands out:

Any treadmill: no smart treadmill, sensors or controllers. The headset alone reads your stride. Real distance: the kilometers you walk carry your story forward. Mixed reality: you see your real room and treadmill, so it’s safe to move fast. Treadlight demo complete ZIP

is this correct : "Treadmill Warrior gamifies exercise and turns treadmill runs into an RPG adventure. Every kilometer walked counts toward a real journey between villages, with quests, levels and a story. "

Checking the game's description for accuracy.

Mostly yes. One part isn’t quite accurate, and one wording choice is worth a thought.

Accurate: the game turns treadmill sessions into an adventure, with quests, XP levels and a story, and villages that are 10 km apart. Not quite accurate: “Every kilometer walked counts.” Only runs that reach their beacon count toward the journey. If you abandon a run, or the darkness takes it, the distance isn’t banked, though you still keep half its XP. To stay precise, you could say “every kilometer of a finished run”. Wording choice: “RPG” fits, since there are levels, XP, quests and a story. The game’s own tagline says “Roguelite AR adventure for treadmills”, though. If you want the pitch to match the title screen, use that phrase. “RPG adventure” is the plainer label for a general audience.

A tightened version:

Treadmill Warrior gamifies exercise and turns treadmill runs into an RPG adventure. Every kilometer you complete counts toward a real journey between villages, with quests, levels and a story.

There’s also a double space after “adventure.” in your draft, which this version removes.

Be sure to write what inspired you, what you learned, how you built your project, and the challenges you faced. Format your story in Markdown, with LaTeX support for math.

Here’s the project story in Markdown, ready to paste into Devpost. Inspiration is the section to make your own: I’ve built it from what the game does, but your personal reason for making it will land better than mine.

markdown

Inspiration

Treadmills are the most effective and the most boring machines in the gym. You walk for forty minutes, staring at a wall or a phone, and the only story is a number counting up. Mixed reality headsets changed one thing that matters here: on a Quest you can see your real room, so it is safe to move fast. I wanted to know what happens if the treadmill becomes the controller and the room becomes the world.

The idea that kept me going: what if every kilometer you walked actually took you somewhere? Not a step counter, but a journey, with villages, a quest-giver who needs your help, and a beacon at the end of every run.

What it does

Treadmill Warrior is a roguelite AR adventure for treadmills, running in the Meta Quest browser with passthrough.

  • Step on any treadmill. There is no smart treadmill, no pairing and no wearables. The headset reads your stride from the small bob of your head.
  • A light trail unrolls through your room and carries you through worlds (the Hollow Wood, Glass Canyon, Dandelion Plains, the Starless Reach) at the speed of your own pace.
  • Play with your hands. Catch wisps of light to fill your lantern, push a palm through shards of gloom, raise a hand to choose your road at forks, and press your palms together to light the beacon at the end.
  • Pip the firefly guides you, steers wisps into reach and dances when the beacon comes into view.
  • A real journey. Every completed run banks its distance toward the next village, 10 km away. The Anthill Village has a stranger with a quest: relight the Dead Beacon beyond Serpent Lake, across a broken trail where the darkness strikes.
  • XP and levels reward distance, strides, wisps, gloom broken and beacons reached, with a bonus for running.

How I built it

The game is written in TypeScript on Meta's Immersive Web SDK (three.js and WebXR), built with Vite and served to the Quest browser. Everything you see is generated in code: no external 3D files are loaded at runtime.

Stride detection from the headset alone. The core problem was speed without treadmill data. When you walk on a belt you stay in place, but your head rises and falls once per footfall. I band-pass the headset's height to isolate that bounce, count a footfall each time it crosses a threshold, and only trust the rhythm after three regular strides. Cadence comes from the mean of the last six intervals:

$$ c = \frac{60}{\bar{t}}, \qquad \bar{t} = \frac{1}{6}\sum_{i=1}^{6} t_i $$

where $t_i$ are the gaps between footfalls in seconds, accepted between 0.22 s and 1.3 s.

Cadence becomes trail speed. A 120 strides-per-minute walk travels at 3 m/s, scaled within limits:

$$ v = 3 \cdot \operatorname{clamp}!\left(\frac{c}{120},\ 0.5,\ 1.6\right)\ \text{m/s} $$

A steady pace instead of a twitchy one. Following cadence directly made the world slow down every time I looked around. So the trail now holds a baseline pace learned in a warm-up, and only adopts a new pace when it stays outside a ±10% band long enough: 4 s to speed up, 8 s to slow down, because a false slow-down feels much worse than a false speed-up.

Progression. XP per run is

$$ \text{XP} = \sum_{\text{events}} b(v)\cdot x_e, \qquad b(v) = \begin{cases} 1 & v < 2.5 \ 1.25 & 2.5 \le v < 3.5 \ 1.5 & v \ge 3.5 \end{cases} $$

with 100 XP per km, 1 per stride, 5 per wisp and 50 per gloom or beacon. Abandoned runs keep half. Level $L$ costs $1000 \cdot 1.25^{\,L-1}$ XP.

Worlds that never repeat. Scenery is instanced, low-poly and procedural: 48 slots recycle along a 72 m window, and each slot re-rolls its prop, position, size and turn every lap, seeded so it changes out of sight. Some worlds are glowing holograms; others (village cabins, crystals, dandelions) are flat-shaded with lighting baked into vertex colors, so dozens of copies stay cheap on a standalone headset.

Interface for sweaty hands. Menus use hold-to-press medallion buttons and a touch tray, so nothing needs a precise pinch mid-run. Story and dialogue text types out letter by letter, with recorded voiceover. Sound effects are synthesized live with the Web Audio API.

Testing without the treadmill. I built a desktop preview that simulates strides, hands and gestures from the keyboard, and ran automated checks in a headless browser: playing whole runs, capturing screenshots and reading the game's state. That let me iterate many times a day and save headset sessions for what only a headset can tell you. I developed the game together with an AI assistant (Claude), which wrote and tested much of the code while I directed the design, art, writing and voice.

Challenges I ran into

  • Reading speed without sensors. A head bob is a few millimeters. Separating footfalls from nods, glances and turning to look at something took a lot of tuning: amplitude thresholds, interval limits and requiring a regular rhythm before trusting it.
  • Looking around shouldn't cost you speed. This led to the baseline-pace system described above.
  • Comfort and safety. The trail has to stay centered on a real treadmill with handrails. Scenery never comes within 2.1 m of the center line, the world can be recentered, and stopping eases the trail to a halt instead of freezing it.
  • Performance on a standalone headset. Every prop is instanced and every glow is baked rather than lit, so whole worlds render in a handful of draw calls.
  • Art direction. Not every model works on a moving trail. Some versions looked wrong in motion, and I rolled them back; the crystal horns, for example, only worked once they became a gateway arching over the trail.
  • Pacing a story around exercise. Distance only counts for completed runs, so a quest needs to fit several real workouts, not one session. The Dead Beacon fills over two to three runs.

Accomplishments that I'm proud of

  • It works on any treadmill with nothing but the headset.
  • Running feels like traveling: villages, a stranger's quest, a Broken Trail that can take your light, and a demo ending at Rybol Village.
  • Pip, who somehow became the heart of the game.

What I learned

  • The body is a great controller, if you design around it. Small, forgiving gestures such as a raised hand or palms together work mid-stride; precise ones don't.
  • Smoothing is a design decision, not just a filter. Players notice a false slow-down far more than a slightly late speed-up, and the math should reflect that asymmetry.
  • Mixed reality is the right medium for fitness. Seeing your room keeps you safe, and the fantasy still lands.
  • Fast iteration needs a fake headset. Simulating the player turned days of testing into minutes.

What's next

The Inner Peace world (a calmer, story-driven place where you recollect the legend of the lost light), more villages and quests beyond Rybol, a Fitness mode with structured workouts, and multiplayer runs.

Some details to check against what you actually did, since they’re your claims to make:

AI assistance: “How I built it” says you built the game with an AI assistant. Keep that line, reword it, or drop it, depending on the hackathon’s rules and your own preference. What’s next: multiplayer and a structured Fitness mode aren’t built yet. The lobby shows Fitness as “Coming soon”, so listing them as next steps is fair, but cut anything you don’t actually plan. Inspiration: the opening paragraph is generic on purpose. If you have a personal story, such as your own treadmill routine or covering XR fitness as a journalist, put it there.

What's next: More levels, quests, worlds, mist world, snow land as well as fitness mode which lets users play to 20 minute mixtapes, and collect wisps in sync with music (a more standard music/fitness game mode)

Here’s the updated “What’s next” section, ready to paste in place of the old one:

markdown

What's next

  • More of the journey: new levels, quests and worlds beyond Rybol Village, including a misty world where the trail fades into fog and a snow land.
  • The Inner Peace world: a calmer, story-driven place where you recollect the legend of the lost light.
  • Fitness mode: a more classic music fitness game. You pick a 20-minute mixtape and run to it, catching wisps in sync with the beat, for a workout you can drop into any time

Built With

Share this project:

Updates

Submission history