Surfr Update

Update 26-07-2023

Leading innovation in height tracking, this time powered by Artificial Intelligence, never was a risk-free path. But the objective is clear, and we just made a huge step into the direction of the end-goal: 100% jump recognition with high accuracy. On 26-07 we took an other big leap towards the finish line.

In this blog we would like to keep you updated on on the latest developments of Surfr.AI. This is a live-blog that gets altered / updated when new information is available. 

Surfr.AI status update 26/07 – One step closer to the finish line!🚀

After the last update of 12/06 we now have released another update to the stores. In this update we trained Surfr.AI more and more to further improve on jump recognition and airtime issues. Please update as soon as possible via the stores to enjoy the new update.

Based on our testing we now see:

99.9% jump recognition  

  • Close to 100% jump recognition – this applies to normal jumps. We are still improving on exceptional jumps such as jumps with multiple heli-loops (although they should go pretty good already by now).
  • Please note: jumps below 1.5 meter are not recorded at all on the Surfr App.
  • Jumps between 1.5 and 4 meter might occasionally be missed as well, the lower the jump the bigger the chance for it to be missed. We know when you first start jumping 1.5 meter feels like serious flying, and sometimes users new to jumping will feel some of their jumps are missed. When we then investigate the session it almost always turns out that those jumps are in the 0-3 meter range. Please know that we are considering a high-sensitivity mode that will pick-up more of those ‘low’ jumps (no offense), but as of this moment ‘low’ jumps  might be missed.

High consistency

It’s one thing to support one platform (such as iPhones) it’s a whole other thing to support all platforms: Apple / Android / Watch OS (apple watch) / Wear OS (galaxy watch).

When developing for multiple platforms there comes a new factor into play: we expect consistency. Taking a Apple Watch and a Galaxy Watch and using it in the same sessions should give similar results. Consistency means that jumps give similar heights, similar airtimes, and that all devices should detect the same jumps.

We have been doing an incredible amount of testing of wetsuit vs board vs watch, apple vs android and sometimes we went with up to 8 devices at the same time onto the water. We compared different ‘modes’ vs different brands. We kept tuning and training Surfr.AI and we finally arrived at consistent results.

Does this mean that every device will give the same results up to centimeters precision? Not exactly. Because height is also affected by what you do. If you raise your arm or kick your board this will increase the height. But on Surfr height tracking is as accurate as it can possibly be. This doesn’t mean centimeter precision. Our goal is to get an difference of less than 5% on 95% of the jumps and a peak difference of not more than 10%. This means that if an Apple Watch give you 10 meter, we expect the Galaxy watch to be somewhere in the range 9.5 to 10.5 meter 95% of the time. In 5% of the time the difference might be a bit bigger, but shouldn’t exceed the 10%, meaning a range of 9 meter to 11 meter in this example. We already achieved our ‘less than 5% difference 95% of the time’ goal , and we’re getting really close to our goal of not exceeding the 10% ever. 

We will keep improving to remain the most accurate height tracking solution across all platforms and devices.  Below an example of the consistency between Apple Watch and iPhone.

Please note that in this picture:

  • On jumps lower than 2 meter one of the devices might miss the jump. This session was a session in the waves in Cape Town, and the sub 2 meter ‘jumps’ are mainly waves.
  • None of the jumps above 4 meter is missed by any of the devices
  • Maximum difference that is recorded on jump 52 is a difference of 1.3 meter. This was a jump where I did a one footer, and therefor lowering my watch arm hence it measured slightly lower. 

This is real data and everybody can reproduced this in similar tests. We encourage people to go out there and put the accuracy and consistency of Surfr.AI to the test. Because; seeing is believing 🙂

Resilient to mid-air movements

Now, let’s talk about a major breakthrough: we have successfully trained Surfr.AI to accurately differentiate between a landing and an abrupt mid-air motion. In earlier Surfr versions, sudden movements like hitting your phone or waving to your friends with your watch could cause jumps to be missed or prematurely aborted, resulting in lower airtime or height measurements. I am happy to inform you that this issue has been resolved.

What’s next

In the coming weeks, we will continue to collect data to further refine Surfr.AI and achieve our goals. If you encounter any issues after updating to the latest version, please send us a direct message on Instagram, and we’ll be more than happy to assist you.

Why the need for Surfr.AI? Wasn’t it all working great already?

Although Surfr was working great already we wanted to achieve higher consistency across different devices / platforms.

The goal of Surfr AI has been to improve on this. Because of the wide variety of sensors that are built into the different devices the job for accuracy and consistency became more and more complicated.

AI offers a solution to complicated problems, such as consistent jump recognition across all platforms. This is why we took the risk to innovate and be the first tracking app that uses AI for jump detection. This turned out to be more complicated than we thought and we had a few road bumps along the way. But we’re happy to have finally arrived at stable and reliable situation of jump detection and accuracy. It’s not 100% perfect yet, and we are doing active monitoring of the leaderboards. In case any wrong data still slips through we will remove it from the leaderboards. In the mean while we will keep training Surfr.AI to get smarter and smarter. 

What is AI and what does Surfr.AI exactly mean?

You may have recently heard about AI, particularly due to the buzz around Chat.GPT. However, AI has been around for a while, finding applications in various fields. At Surfr, we are now incorporating AI into our algorithms to achieve higher accuracy, improved jump recognition, and consistent performance across devices.

For those who are interested, let me briefly explain what we are doing and what Surfr.AI truly represents. To understand this, let’s imagine the human brain with its remarkable ability to learn, remember, adapt, and improve.

Now, let’s imagine that a big-air competition organizer approached you to be a judge in a kiteboarding event, but you have no prior knowledge of megaloops, back-rolls, double loops, kiteloop-boardoffs, boogieloops, and so on. This sport evolves rapidly, with new tricks being invented almost every week.

How would you learn to differentiate between all these different tricks and become a competent judge? The answer lies in observing numerous examples. The more examples you see, the more “trained” you become, and the better judge you will be.

Surfr.AI works on the same principle. It employs neural networks that mimic the human brain’s ability to recognize various types of jumps. Each jump is unique and differs from the others. We have floaty 20-second jumps with multiple heli-loops, low jumps, high jumps, hard landings, soft landings, and so much more. Surfr.AI can record all of these jumps, whether you’re using a phone, a watch, on your board, wrist or in your wetsuit.

One advantage of a digital brain compared to the human brain is that AI never forgets what it has learned. Let’s go back to the big-air judging example: if a judge takes a break from judging, they may forget some tricks and struggle to recognize them later. Fortunately, a digital brain doesn’t face this problem; it retains everything it has learned.

So, this is what we have been working on for months and what we mean when we mention training Surfr.AI. We have trained Surfr.AI with thousands of examples based on various kite sessions. In recent weeks, we have gathered additional data from several Surfr users (with their consent) and used it to train Surfr.AI.

Just like the learning process in the human brain, training Surfr.AI is an ongoing endeavor. We will continue to collect data and train Surfr.AI to make it even better. However, this release marks a significant step in the right direction. We are eager to hear your thoughts on how Surfr.AI is performing for you. Please share your feedback via Instagram DM, both when it’s working perfectly and when it’s not yet perfect. The latter is temporary because, with our fast-learning AI as the foundation of the Surfr algorithms, it’s only a matter of time before perfection is achieved.

Thanks for flying with us, regards,

Herbert – Founder of the Surfr App, and the rest of the team!