Power Up Your Pedals

CycloWatt, a low-cost IoT device, utilizes edge machine learning for accurate cycling power measurements to enhance training sessions.

Nick Bild
28 days agoInternet of Things

Cycling power meters have forever transformed the way cyclists train and analyze their performance. These devices are sophisticated tools designed to measure the power output of a cyclist, typically in watts, allowing for precise monitoring of effort during rides. They are usually attached to the bicycle's crankset, pedal, hub, or sometimes integrated into the bike's frame itself. Once installed, they continuously measure the force applied to the pedals and transmit this data to a cycling computer or smartphone app for real-time analysis.

One of the primary purposes of cycling power meters is to enhance training effectiveness. By providing accurate and instant feedback on power output, cyclists can tailor their workouts more precisely to their training goals. This data allows for the creation of structured training plans that target specific power zones, helping cyclists optimize their performance and improve their fitness levels over time. Additionally, power meters enable cyclists to track their progress more accurately, identifying strengths and weaknesses to focus on during training.

These devices are also widely used to prevent injuries that can result from overexertion, and to aid in muscular development. But despite their obvious utility, cycling power meters have some drawbacks that have limited their widespread adoption. Cost is a significant barrier for many cyclists, as high-quality power meters can be quite expensive. Additionally, the process of moving power meters between bikes can be complex and time-consuming, making them impractical for cyclists who regularly switch between multiple bikes.

A trio of engineers at ETH Zurich have taken a novel approach that reduces cycling power meters down to a small and inexpensive Internet of Things device called CycloWatt that can easily be moved between bicycles. They were able to achieve this feat by leveraging edge machine learning techniques in conjunction with low-power sensing technologies. The team’s meter proved to be nearly as accurate as much more expensive commercial options, and it can run for nearly 26 hours on a single battery charge, adding to its practicality.

Rather than being permanently installed on a bike, CycloWatt takes the form of a cleat, which can be installed on a cycling shoe by simply attaching it with a few screws. Inside of the cleat, the device contains a load cell for measuring forces of up to 1,000 newtons. This load cell works in conjunction with an inertial measurement unit (IMU) to feed measurements into a STM32L4 microcontroller equipped with an Arm Cortex-M4 processor. A low-power Bluetooth module is also integrated into CycloWatt’s custom PCB to facilitate the real-time wireless communication of relevant information. The total cost of the components is about 70 dollars.

In order to turn the load cell and IMU readings into a measurement of power, the team designed a compact neural network that could run within the tight resource constraints of the chosen microcontroller. After training this network consisting of 70,337 trainable parameters and 546 neurons, it was observed in a series of experiments that the mean absolute error was only 12.29 watts, or about 4.1 percent. This was determined by comparing the predictions made by the model with ground-truth power data captured from a commercial Stages Cycling R7000 power meter.

Looking ahead, the team plans to tweak the mechanical setup of their custom cleats to prepare them for use by a wider audience. They also intend to explore the possibility of training their model on a larger dataset, which could further enhance the system’s accuracy.

Nick Bild
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles