MotionPlay | Devpost

Inspiration

Developers and programmers frequently endure long hours glued to their chairs, contributing to a highly sedentary lifestyle with serious health consequences. Prolonged sitting is linked to increased risks of obesity, metabolic syndrome (including high blood pressure, elevated blood sugar, excess waist fat, and abnormal cholesterol), cardiovascular disease, type 2 diabetes, certain cancers, back and neck pain, musculoskeletal disorders, and even higher overall mortality—comparable to the dangers of smoking or obesity when exceeding 8 hours daily without breaks or activity. Sources like Mayo Clinic highlight that sitting more than 8 hours a day with minimal movement elevates death risk significantly, independent of exercise elsewhere. We created this project to fight the "sitting epidemic" by turning mini-games into enjoyable bursts of motion right at their desk—using AI-driven, controller-free gameplay that prompts natural movements like jumping or subtle gestures. This makes screen time healthier, injecting fun, light exercise into coding marathons to improve posture, boost circulation, and reduce sedentary-related strain without leaving the workspace.

What it does

Mini Game Project is an endless runner-style jumping game where TensorFlow.js AI automatically detects hurdles and triggers precise jumps for the character—no manual controls required. Players experience thrilling, autonomous gameplay while the setup encourages standing, light jumping in place, or body engagement to stay active. It's ideal for quick desk breaks, blending entertainment with gentle physical activity to counteract prolonged sitting's effects like stiffness, poor circulation, and fatigue.

How we built it

We forked an open-source JavaScript endless runner and supercharged it with TensorFlow.js for real-time machine learning. A lightweight neural network processes canvas frames to predict hurdle approaches, outputting jump commands with minimal latency. We optimized the model for browser efficiency, integrated inference into the game loop, and ensured smooth animations across devices.

Challenges we ran into

Timing jumps accurately was challenging—tiny delays from model inference caused frequent collisions, requiring extensive training on varied scenarios for better precision. Balancing AI computation with animation performance in browsers demanded optimization to avoid lag. Collecting diverse data without overfitting, plus ensuring cross-device fluidity, tested us under tight hackathon deadlines. Having a working demo

ACHIVEMENTS

We're proud of a smooth, demo-ready game where the AI achieves over 90% jump accuracy across speeds and patterns, making the experience feel intelligent and effortless. Delivering this innovative, health-focused twist on a classic game in limited time is a standout win.

What we learned

We honed skills in repurposing open-source code to accelerate development, focusing energy on AI integration rather than rebuilding basics. We also mastered TensorFlow.js for web-based real-time ML—handling model training, optimization, inference speed, and browser constraints effectively.

What's next for Mini Game Project

We plan user-friendly customization: easy spritesheet uploads and character swaps to spark community creativity. We'll add more mini-games with gesture/motion controls for varied movement (e.g., dodging or reaching). Exploring reinforcement learning could create adaptive AI for escalating challenge, plus multiplayer human-vs-AI races. Ultimately, we envision a collection of fun, movement-promoting tools tailored for desk-bound workers to stay healthier during long sessions.