Elective — 2018
Smart Training Glove

About the Concept

The Smart Training Glove is a tool that is able to provide constructive strength training feedback without invasive sensors or difficult set-up procedures. The gloves analyse movements made in real-time, detecting which exercise is being done and how well it is performed. The machine-learning powered back-end keeps track of the exercises and training schedule, while the front-end displays graphic constructive feedback.

In collaboration with Melvin Sterk and Shen Yiwen.

Instantaneous Feedback

When performing strength training exercises, maintaining good form is essential [ 1 ]. Bad form does not only heavily hinder muscle growth and caloric burn rates, it can also lead to nasty short- and long-term injuries.

The Smart Training Gloves are able to detect deviations from perfect form and show easy-to-understand graphic feedback so you can improve your training form during the workout.

Image Title
Image Title

Learning from Data

Triaxial acceleration data captured during exercises is evaluated with the trained machine learning model and placed in one of the pre-specified classes. The features extracted from the data stream are the arithmetic mean, the standard deviation, the minimum and the maximum for the X-, Y- and Z axes.

The training model was created using a Linear Support Vector Machine [ 2 ], fed with a self-recorded training data pool of 361 exercises, recorded by multiple users. A 5-Fold Cross Validation was implemented to generate a more robust model, and a C-parameter value of 20 was found to yield the highest in-sample and out-of-sample accuracy. After optimizing, an accuracy of 98.04% was achieved!

Companion App

Alongside the physical product, a companion application was designed for both mobile and computer use. This app is packed with custom-designed graphics, and offers both exploratory- and informative content.

Explore a collection of exercises, create workout schedules tailored to your needs, watch instructional videos and read step-by-step instructions to achieve better results from your fitness workouts.

Image Title

Conclusion

This project was graded 8.0 out of 10. The course has brought me a valuable new tool in adding extra depth to product designs by implementing machine intelligence. Throughout the curriculum I have learnt about the different types of learning — Supervised-, Unsupervised- and Reinforcement Learning — and improved my intuition regarding common machine learning algorithms. Something that I am very eager to do is to implement machine learning on data that is not based on physical sensors, such as website / application usage.

My vision on implementing machine learning in design is to provide mass customization of users’ experiences, environment and system responses. One of my favourite commercial implementations of machine learning is Spotify’s Discover Weekly: their algorithms are able to predict with incredible accuracy as to what music I would enjoy, based on my previous usage. I think this is absolutely brilliant: they have managed to create a product that distinguishes itself from other music streaming applications by creating personalized experiences for all its users.

References

  1. Westcott, W. (2012). "Resistance training is medicine: effects of strength training on health.". Current Sports Medicine Reports 11(4), 2012. Retrieved Feb. 6, 2018, from Pubmed.gov
  2. Chang, C., Lin, C. (2016). "LIBSVM -- A Library for Support Vector Machines" Retrieved Feb. 6, 2018, from NTU CSIE