Gyrofit

GyroFit

Serious Game  /  Machine Learning  /  Unity  /  MSc Dissertation 2024

GyroFit is a mobile serious game I built for my MSc dissertation at UWE Bristol. It uses the phone gyroscope to track physical exercises in real time while a machine learning model generates a personalised daily workout plan based on the user’s profile. The avatar mirrors the player’s movements and changes appearance as they progress. It worked. But it also made the limitation obvious: the system adapted to what users entered at the start, not to how they were actually doing week by week. That gap is what I am still thinking about.

Grade: 85    Supervisor: Dr Nikolaos Ersotelos, UWE Bristol

Research Contribution

Technical Stack

  • Unity (C#)
  • TensorFlow — ML model training
  • ONNX — model deployment inside Unity
  • Gyroscope sensor input via mobile
  • Blender, Substance Painter

Key Findings and Limitations

  • ML pipeline from TensorFlow to Unity via ONNX is viable and performant on mobile
  • Sensor-based interaction creates meaningful engagement in exercise contexts
  • Static user profiling is insufficient for sustained motivation
  • Dynamic adaptation based on in-session behaviour remains an open problem
  • Avatar feedback as a motivational signal warrants further investigation

Screenshots

Intro Menu UI

  • Name
  • Gender
  • Age
  • Weight and Target Weight
  • Main Menu
  • Exercise Menu
  • Live Data
  • Success

Demo Video