{
  "id": 4688092,
  "title": "AI wants to turn a simple phone video into cycling performance data serious riders pay thousands for",
  "url": "https://urgent.news/2026/08/31/ai-wants-to-turn-a-simple-phone-video-into-cycling-performance-data",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-31T18:15:33.000Z",
  "source": {
    "name": "Digital Trends",
    "slug": "digital-trends",
    "url": "https://www.digitaltrends.com/cool-tech/ai-wants-to-turn-a-simple-phone-video-into-serious-cycling-performance-data/"
  },
  "original_language": "en",
  "account": "Cycling enthusiasts often invest significant sums to measure and optimize their performance. In a professional setting, this may involve instrumented pedals, bike computers, heart-rate sensors, and more, all aimed at extracting maximum information from each ride. However, researchers at La Trobe University are now investigating whether a device that most people already possess – their smartphones – could play a crucial role in this process. Specifically, they are developing AI models that can estimate the forces applied to a cyclist's pedals simply by analyzing a video recording taken with a phone.\n\nThe team at the Holsworth Biomedical Research Centre is leveraging deep learning to establish a connection between the visual cues observed in a cyclist's motion and the actual forces detected at the pedals. To achieve this, the researchers trained their model using synchronized laboratory recordings, which included one dataset consisting of video recordings of the rider in motion and another dataset measuring the real forces applied during each pedal stroke. By learning the relationship between these two sets of data, the AI aims to predict the forces based on movement alone.\n\nEarly results from a study conducted by La Trobe researcher Rodrigo Bini using recurrent neural networks have shown encouraging outcomes. The model was able to predict three-dimensional pedal forces as well as forces and power at lower-limb joints, with correlations ranging from 0.79 to 0.96. While accuracy varied depending on the specific measurement being estimated, this research has the potential to make such analysis more accessible outside the laboratory setting.\n\nAt present, the research is still in its early stages, and La Trobe does not plan to release a smartphone app anytime soon. However, Bini estimates that within the next five to ten years, smartphone apps could enable almost anyone to record their movement and estimate cycling forces. The team is currently working on strengthening and validating their dataset, with preliminary findings expected to be presented at the International Society of Biomechanics Conference next year.\n\nWhile dedicated smart pedals, such as the Garmin Vector 3, can provide detailed information about force generation during pedal strokes, this level of granularity has traditionally required specialized sensors and substantial financial investment. By turning a smartphone camera into an approximation of this toolkit, the researchers believe that genuine cycling biomechanics could become more readily available to a wider range of riders. As AI continues to be integrated into various aspects of our lives, such as fitness coaching, we can expect to see further advancements in fine-tuning physical aspects of our well-being.",
  "summary": "Researchers are developing AI that can estimate cycling pedal forces from video, potentially turning ordinary smartphone footage into useful performance and injury-risk data.",
  "key_points": [
    "La Trobe University researchers develop AI to estimate cycling forces from smartphone videos",
    "Model predicts 3D pedal forces with correlations of 0.79 to 0.96",
    "Smartphone app could make cycling biomechanics accessible to wider range of riders"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}