24-February-2020 By Jeffrey Cooper

The Connected Gym: Tier 3- Sensor Fusion

In the previous articles Connected Gym, Part 1 and Part 2, we examined how to outfit your club with a system to allow automatic capture of workouts in the club using a smart watch and beacons. Part 1 described a minimum setup that has a clear set of benefits. Part 2 expanded that across the club, applying the same concept and delivering a better set of analytics and resulting benefits.

This- the last of the technical articles in this series on the Connected Gym, will focus on an advanced system and additional benefits. You will need to implement Part 1 and Part 2 to implement Part 3.

With data from multiple sources- beacons, connected equipment (when available) and sensors on the watch, you will have the full context available. Context is what allows you to make sense of what is happening.

Going Deep to Create the Fully Connected Gym

Once you’ve established automated check-in and check-out, and broadened coverage to the rest of the club, you will be getting reams of analytical data on club usage and member habits. It will take a while to adapt and learn how to make use of that data, so I don’t recommend jumping to this level until you’re ready. At this point, you have a partially connected gym. However, it’s a good idea if you have, to keep the data until a future date that you can analyze it.

This deeper level involves using everything at your disposal to improve the granularity . As an analogy, consider how telescopes have improved over the decades.

This series shows the famous Crab Nebula, as seen in a backyard telescope. This is akin to the current state in most clubs- check-in data plus some odds and ends. The next is a photo from the best telescope on Earth prior to the Hubble launch and the new era of deformable-mirror telescopes. It shows much more detail, but it’s still lacking in the interior. This would be similar to Part 1 of this series and give or take, Part 2. We can tell a lot from this and it got us a long way to understanding these structures.

The best image is from Hubble. You can see so much more structure. With data like this in multiple wavelengths blocked by Earth’s atmosphere, scientists can sa lot more about this nebula and on nebulae in general. This article is looking for that type of clarity in your club data with a fully-connected gym.

Up until now, the Smart Watch mostly has been doing it’s thing- measuring heart rate, and reading beacons. The beacon detection is a relatively low-frequency event- it does not need to happen constantly. Ideally, the watch keeps tabs on the location and only makes changes when it detects a room or region change (say, cardio to weight room).

In this article, we’ll explore using everything the watch has, which is accelerometers, gyros, and additional sensors in the future. We’ll also be able to fold in external sensors, when and where available. Let’s go through the various areas of a club where this will make a difference.

Cardio

We mentioned in the previous article that not all clubs have the latest and greatest cardio equipment. It’s expensive, for starters, and they generally have an approximate lifetime of 7 years in a club. Apple’s GymKit has not been around that long, and indeed, gym-capable wearables have not been around that long.

My own club has quite a few human-powered cardio machines- they do not even plug in. They simply rely on the exerciser to start, and the device generates its own power. Stop, and the display goes out.

Using motion analysis from the smart watch’s accelerometers is an easy way to disambiguate which machine you are on. A treadmill, elliptical, cycle, rowing machine and step machine all generate different motion signatures from the wrist perspective.

In an example above from an academic paper, you can see, rather dramatically, how the motion signatures differ from exercise to exercise. The algorithms do not need to look at the entire spectrum of possibilities. According to the beacons, the user is in the Cardio room. The club lists, for this location, 4 types of machines, for example. The algorithms, thus constrained, only need to look for four patterns, with some variability, naturally, from person to person and different brands of smart watches.

Later versions of smart watches today can auto detect these workouts. The issue with that is it is internal to the watch and is not information that is easily transferable to a club or unified into a single report to the member. (It’s the same with swimming, running, and cycling automatic detection.) It is inside the manufacturer’s walled garden of data. At best it can be skimmed from the phone’s health app after everything is done, but it won’t be real time within the club app.

Weightlifting

The same idea applies to weightlifting. The number of options is higher as there are many more ways to lift weights than there are types of cardio machines, so it is more difficult. But, it’s doable and has been demonstrated on a number of different brands of smart watches.

It’s possible to identify the actual movements the member is doing. Similar to the cardio accelerometer examples above, each movement generates a unique signature in those graphs. With that, you can detect bench press, deadlift, bicep curls, squats, etc…

Once you can identify most movements, you can comprehensively show the member what muscles they are working out, or collectively, which muscles they worked out today and how effective it was. This feedback is important for helping members understand what their workouts are targeting and how they are helping.

It’s similar with selectorizor (cable & pin) machines as well, and in fact, it’s a bit easier because the motions are constrained and not completely free form. Clubs generally also separate selectorizors and free weights, so it’s easy to tell the difference by reading the beacons.

Advanced Weightlifting

As advanced as this is, you can take this a step further. For selectorizors, there are several solutions already that can help identify who is at the machine and how much weight they are lifting. They can either do this via a sensor clamped to the cable or via visual recognition. Systems such as ShapeLog and FitTrack

For free weights, it’s a little more complicated. Possible solutions involve individual sensors on the weights, though this has proven complicated (and all those batteries!). The other potential is visual recognition, which will take some fine tuning to tell apart, but at least you don’t have 300 sensors in your weight room all emitting radio signals. In time, this will be solved, but at least in the short term, I don’t know of any deployed solutions.

Cross Training

Cross Training is complex. In theory, you can detect all individual parts of a person’s cross training workout. I cross train, and do interleaved sets of reps in clusters. For example, I’ll interleave sets of battle ropes, kettle bell on a BOSU and platform jumps. Then it’s ab bench, jumps holding weights, slam ball and high-kneed sprinting in place. Then jump rope and more footwork (for trail running support). Standard exercises such as battle ropes and slam ball should be easy, but several of these are my own inventions, and are intended to improve my stability and balance when I run trails, and cannot easily be quantified.

Reporting such a mixed and sometimes free form workout would be confusing to interpret at best. While you can clearly break down a workout by groups- cardio, weightlifting, and it’s common to break down weight lifting into the various, standard movements. Cross training defies classification and is decidedly messy A little intelligence in the app could simply classify this part of your workout as Cross Training, while elaborating weightlifting.

Personal Training

Personal Training may gain the most from an advanced data gathering capability. During a personal training session, the trainer can just tell the member what to do next. The watch can simply feed back the heart rate, which can guide the trainer on your intensity.

The real benefit, though, is the trainer can push custom workouts to each member’s phone and smart watch, tailored precisely to that member, to do on days between training. When the member comes in, they can execute that custom, coached workout, and the watch will record it and report it back to the trainer. Pear Sports is a great example of a company that is already doing this utilizing both phones and smart watches.

The next time you show up for personal training, the trainer can review those workouts and evaluate them. This is a capability that can allow you to increase your offerings to your members.

Personal Training is expensive. I paid $3000 for 3 months of training in 2005 at Life Time Fitness, and the cost will be much more today. It was the best investment in my health I ever made, and it still pays off because it locked me into fitness for life. My trainer was excellent, challenging and a great teacher. But that’s expensive for many people, and indeed was expensive for me at the time. While some can stretch and pay for it, some simply cannot.

Clubs can offer different levels of personal training at lower prices. Mine was 3x a week. Perhaps you can offer once a week with automated (but still custom) workouts sent to the member for twice more that week. So you have one real coached training and two virtual (with a human touch) coached workouts. Perhaps you offer it at 40% the full training price since your trainer is used less. Now you have a tiered offering.

You can also offer special boot camps, at a lower cost, and push these out to amenable groups.

Group Fitness and Other Areas of the Club

For group fitness, I would need to study this more to see if additional data could help. The room + schedule will clearly identify you are taking a Barre class, Zumba or LesMills Body Pump. And heart rate clearly is correlated- directly, with workout intensity. I’m unsure that more data is needed, and while data is golden, unneeded data can be a burden. If any reader has thought this through, I would welcome the input.

It’s similar in the remaining areas. Rock climbing is a permanent facility if the club has it. In larger clubs with basketball courts, accelerometers could differentiate between basketball and volleyball, for example.

Watches can auto-detect stroke types and laps swum, but are siloed in their walled garden until later. Notably, there is the excellent Swim.com application, which is the best in class, but it is still independent. I recommend you use it. Perhaps one day the algorithms might be white-labeled into the club app so everything is in one place.

Conclusion

Connected Gym Parts 1, 2 and 3 together take a deeper and deeper look into outfitting your club with advanced analytics to help both you and your members. Your members will get more detailed reporting, and importantly, correlate their activities with the results. Are they losing weight, is there resting heart rate going down, are they sleeping better, are stress levels lower?

With a fully-connected gym, you, the club, will better understand each member’s journey and be able to help them better achieve their goals. You will learn to more easily identify which members are at risk for dropping out and help them “get over the hump”. You’ll be able to more easily identify those that would truly benefit from the help of a personal trainer, and if they can afford it, it’s an option you can offer to them.

The next article will explain the benefit to the members, keeping it personal and human, and the benefits to the club from the analytical data and where it affects the bottom line.