I saw Shred’s original “data science” thread on needing help with R, and it got me thinking that there are probably quite a few data scientists roaming the trails of the UK.

A bit of background info:
A friend and I are building a wee piece of tech which we’re calling Salvo. In short it’s a small device loaded with accelerometers, gyroscopes, a barometer and a magnetometer. It is designed to measure trail data and quantify mountainbiking beyond simple GPS A – B times, we’re aiming for full on 3D mapping of ride intensity & skill. We’ve got the hardware dialled, but we need help building an algorithm to sort through the ride data.

For the past couple of months we’ve been talking to data scientists/ analysts, hoping to find someone that likes the project enough to help us out. Problem is that none of them were a mountain biker, so they didn’t really see the point.

So, anyone here want to have a chat about MTB metrics? Who knows, if we pull this off it could be the Strava for mountain biking!

Sounds great but wish I had more experience to help out.

Good luck with the project.

Not really “data science” as the usual commercial use of the term. More basic analysis.

Anyway (end pedant) happy to talk, three maths degrees, over years of programming experience, in mathematical modeling and machine control. Email in profile.

I can ask across our company’s customer analytics team. It is huge and grows month on month, surely by statistical chance one of them must be an avid MTBer.

surely by statistical chance one of them must be an avid MTBer

Even if not they should be able to give an indication of when one will be along. 🙂

P.s. how have you made your muC choice without an algorithm to to get an idea of the complexity and hence requirements? Or have you just gone with something quite powerful?

I think you need to have a think about what your metrics are, it may not necessarily be so much of a a computing issue as problem specification. What do you mean by skill? Can you express that in terms of your measurements?

I may not be your ideal programming guy but I do know a decent amount about maths and algorithm design (my phd) and research in general (my career).

Problem is that none of them were a mountain biker, so they didn’t really see the point.

To an extent, as a biker, I don’t really see the point either. Some existing GPS devices have barometers to aid with height data, so their tracking is already 3D.

I think you need to have a think about what your metrics are, it may not necessarily be so much of a a computing issue as problem specification.

My thoughts as well. Over and above what is generally off the shelf, what is the extra data going to tell a user that is significantly helpful to them or their improvement.

I get that big data capture can tell you more about something. I get that in the case of the bike, knowing lots of info on its position, angle, speed, etc could tell me down to a very accurate scale where the bike may be on a trail at a particular time . . . but what do you do with it?

Am I going to learn about improving my fitness/riding by knowing I took a corner a few inches to the left or right, or that the bike was leaned over a few degrees more than the last time? How could I apply that knowledge to the next ride “this corner’s coming up, I know I leaned to 30deg last time, i’ll lean to 32deg this time…”

Or am I going to learn more knowing from GPS A>B that I nailed a mile long climb by taking 20 seconds out of it and that my heart rate stayed the same?

I’m not trying to be arsey about it, but they are honest thoughts about the comment of problem specification.

Dabbling in data science now as a sideline and making the migration from matlab to Python. And signed up data/metrics/GIS nerd. Not sure it’s the right skill set but I’m certainly interested.

Fisha – gravity races are won by fractions of seconds. Knowing where you lost those fractions could be crucial. Or like me you have disposable income and a love of graphs…

Have you seen the Lit pro its a similar thing

Gravity races may be won by small margins, but do you think competitors would share data such that it would aid someone else in a race at that time? Unlikely. So how would you know where you lost out? It may well be uploaded later, but by then the moment is gone and another day would make conditions different surely. Also, is it not possible that by focusing on the tiny detail, you lose track of the bigger picture… I.e concentrating on one corner at the expense of a longer better line overall. I get that marginal gains can all add up, but they are possibly more suited to gains in repeatable conditions. Can repeatable conditions really be applied to a gravity race?

I know I’m sounding down on it, but I’m just trying to understand what you learn from the data that you can achieve a gain from?

a four axis inclinometer would be ace as well. Bloke at college (Boksenburg you still around !!) built one for motorcycles which was great fun.

I’ve unfortunately got no expertise that can help, but I do have a question: why the hardware? Many mobile phones have accelerometers, gyroscopes etc built in already, so why aren’t you just building an app?

I do agree that more accurate measurements of bike angle (fore-aft during a climb for example) as well as altitude would be good for performance analysis, but I don’t see what would stop Strava for example adding it to their app using existing smartphone sensors if there was a market for it.

Damn, so many replies and some good points made here. I’ll answer them all!

TheBrick

three maths degrees, over years of programming experience, in mathematical modeling and machine control

I should have posted in this forum months ago!

how have you made your muC choice without an algorithm?

We kind of went for the “powerful” option. The device is designed to be a bit future proof, so the quality of the readings are beyond what we need. Having said that we’re co-developing it with another company who has experience in creating these kinds of devices, talking with their data guy to ensure it can do the job. We’re fairly confident that it will deliver the data we need to tease out the information we want.

I’ve sent you an email with more details
———————————————————————————————————————-

GJP

I can ask across our company’s customer analytics team. It is huge and grows month on month, surely by statistical chance one of them must be an avid MTBer.

That would be great, if any of them have a background in time-frequency signal analysis it would be awesome to hear from them. Even if they don’t feel like taking part, anyone’s point of view on this helps us shape Salvo 🙂
———————————————————————————————————————-

thecaptain

I think you need to have a think about what your metrics are, it may not necessarily be so much of a computing issue as problem specification. What do you mean by skill? Can you express that in terms of your measurements?

You’re absolutely correct. We’ve been thinking long and hard about what it is that riders would want to know and we have a couple of different specifications drawn up. At the moment we’re still in early stages and refining Salvo’s MVP (minimum viable product) is key to everything else we do.

If you want I can send you an email with more info. Or if you don’t want your email address on the forum simply send me an email – simon((at))salvo.io
———————————————————————————————————————-

fisha

I’m not trying to be arsey about it, but they are honest thoughts about the comment of problem specification.

Thanks for taking the time to write all that down! Believe me when I say that we need good criticism like this, picking flaws in an idea is what allows it to evolve. Salvo’s specification is still somewhat malleable as we get more input from riders and questionnaire responses.

At the moment we’re still on the fence about how much of a training tool we want this to be for exactly the reasons you mentioned above. Initially I think we’ll have to limit training to basic skills – learning to bunnyhop for example. We could do more, but as you said, with the constantly changing terrain the benefits for an advanced rider are very questionable.

I believe Salvo’s main attraction will be its SSX (that EA game from years ago) style metrics. Where you get an overall score of ride intensity that you can dig into, giving you stats on air-time, jump height, cornering lean angle, crash G-force…. Cool things you can be proud of doing but have no other way of knowing you did. You could overlay these stats on your action-cam footage for a cool F1 style POV. Taking the gaming analogy further, these sorts of stats could also be used in online local, national and global virtual tournaments….ever wanted to know who the airtime ninja in your area is?

With regards to GPS, well, they’re all a bit sh*t when it comes to valleys and dense forestry. One of the main things we’re trying to do here is build a device that does not rely on external systems such as GPS. No data drop out and far higher data resolution.

If you want I’m more than happy to ping you an email or hash more of this out here 😀
———————————————————————————————————————-

jambo

Have you seen the Lit pro it’s a similar thing

LitPro is a great example of innovative tech in extreme sports, I’ll admit when I first discovered them last year I got a bit distressed. But Salvo is quite different in it’s focus and execution, we want to make something everyone can use on any trail….even trails that don’t exist. Not just Pros riding the same loop 50 times.

I’ve sent an email with some more info
———————————————————————————————————————-

jimoiseau

why the hardware? Many mobile phones have accelerometers, gyroscopes etc built in already, so why aren’t you just building an app?

Good question. The problem is all to do with the location of the phone and the calibration of the device for the depth of data we’re looking to get.

Basically we need the device to sit either next to the rear axle mounted on disc mounts or around the BB (deciding which will depend on the data me chose to prioritise, it could be that we do both). Accurate placement is key if you want to be able to compare results with others, a couple mm of difference in placement can have a big effect on the data it reads out.

This “mechanical calibration” is going to be key to making this work as digital calibration would require us to have a database of all frame geometries past and present, a bit of a faff to say the least.

Also, realistically no one wants to attach a phone to their disc mounts!

Interesting, and I have some relevant expertise (I work in robotics and machine learning), but, despite having a good idea of what’s possible, I still struggle with the why. I get why the stats and the gamification might be cool, plus the “better strava” in that rather than the hideous inaccuracy of gps giving you +/- potentially 10s of seconds on DH trails where you care about fractions, but…. is there really a market?

I can see why the above might come across hostile, it’s not intended to – I’m a big believer in trying something because you can, or its cool, or even just in scratching your own itch. Seems a bit like you’re working on the “how” though before the “what”. Have you considered making a prototype with all the functionality and backend as an app? I know the data will be worse, but to work out if it’s even fun/useful data before you try and make it better. Or is it just down to “strava is fun, more strava!”?

Oh and I have thought about doing this before actually, using a pebble for the sensors strapped to a fork leg or similar. I think mounted in the right places you could get interesting suspension telemetry.

Thought about a pebble for a proto? Gets you btle accelerometers in a waterproof package you can easily mount, cheaply

Have you seen the skiing version, PIQ? I love the idea of a mountain bike version just because I’m a sucker for a gadget, but I can’t really think of what useful data a rider could get out of it tbh. I’d like something I could tape to my fork leg or rear triangle and then it would tell me how I should set damping/rebound etc, but knowing I pulled 1.5g in a berm and my biggest jump was 1.2 seconds long isn’t that interesting.

small device loaded with accelerometers, gyroscopes, a barometer and a magnetometer.

This may be a totally off the wall suggestion, but

could you fit it inside the head of a golf club… or even a ball?

same would apply to tennis, cricket, football of course – speed, force and acceleration would all be vital figures for training and performance development.

If you don’t know what the ultimate goal is here, it’s going to be quite difficult to design an algorithm to measure any relevant statistics i think.

Really, you need to start pulling lots and lots of data out of it with preferably with a very skilled rider, and some “less skilled”, to try and see if you can analyse the data to find patterns. This is probably where you need someone with statistics knowledge to come in and help. These patterns will be useful in designing the MTB metrics.

You need to get the riders to do the things you want to measure. Ride fast, ride smooth, get lots of air time etc. then look at the data.

As with others though.. I am a little skeptical. Maybe it’s useful for sports commentators?? similar to the “possession” stats they show for football?

I work with various computer vision algorithms, and I have wondered about using something like SFM (Structure from Motion) or SLAM (simultaneous localisation and mapping) to try and improve GPS accuracy for MTB, but it’s quite labour intensive to develop.

Microsoft have done something slightly similar with their hyperlapse software which smooths out first-person videos
http://research.microsoft.com/en-us/um/redmond/projects/hyperlapse/

The couple of people mentioning suspension set-up/telemetry above, just checking you were aware of this;
http://sussmybike.com/

It’s not quite the same thing really but the Powerpod power meter includes most of the sensors you’re talking about and when combined with their ISAAC software can give braking points, braking forces etc etc. The software gives really in-depth analysis and is free to download so might be worth a look to see if you can use it to get started.

On the down side though the depth of analysis possible means it’s quite complicated and for the recreational user it’s not actually that useful. Obviously if you’re trying to shave seconds from a pb then you can properly dig down into the weeds. To make yours work for the Strava market you’d need to simplify it a fair bit in order to show the results at the trail if not give live updates.

Can repeatable conditions really be applied to a gravity any offroad bike race?

No, which means detailed analysis is pointless for 95% of people. In fact you can still learn plenty from a basic stopwatch.

overall score of ride intensity

You need to define it in terms of actual numbers, good luck 😯

With regards to GPS, well, they’re all a bit sh*t when it comes to valleys and dense forestry.

Not true, although the typical running / cycling ones aren’t the best, there are more accurate ones out there.

Here you go:

http://www.lancaster.ac.uk/dsi/education/business-engagement/

We are looking for partner organisations to offer business-relevant challenges for our students to address over a 12-week full or part time placement, during which the student will work on their specified project. Partner businesses can also benefit from our activities programme.
A partnership with Lancaster University’s Data Science programme offers organisations access to the latest analytical technologies and techniques, backed by a global top 1% university.
You have the data – a Data Scientist can bring you the knowledge.

No, which means detailed analysis is pointless for 95% of people. In fact you can still learn plenty from a basic stopwatch.

This doesn’t really matter. It sounds very cool to visualise and share representations of a ride and compare to my mates.

For that reason people will buy it. Do a kickstarter….

I’m loving the feedback you guys (and gals? It’s hard to tell from the names) are giving me 🙂
It seems like the “why” of salvo is a common question, so i’ll try to clarify.

IA

I get why the stats and the gamification might be cool, plus the “better strava” in that rather than the hideous inaccuracy of gps giving you +/- potentially 10s of seconds on DH trails where you care about fractions, but…. is there really a market?

From what we’ve seen in our market research, yes. To summarise it, the combination of a growing MTB market and the explosion of the quantified self makes for fertile ground for Salvo, but I agree that it won’t be for all riders and like any new idea there is a level of risk. So that is the wholly uninspiring and very top-level business answer which VCs will want to hear, but that isn’t the reason we started this.

For us the “why” of Salvo is quite simple. James and I have been riding mountain bikes our entire lives, I used to race XC in Scotland and James dabbles in Enduro and DH racing (side note: we’re both doing the Ard’ Rock Enduro this year, should get some good data from that 😀 ). We grew up with computer games, Mobile phones, the internet etc… and as a result we love tech just like we love riding our bikes. Yet there isn’t anything that really brings the two together.

Action cams are the closest thing to being able to bring your MTB experiences to the internet but realistically they do a poor job of it. The whole process of filming, editing and then uploading is a faff, and then you end up with an underwhelming video that makes it all look so easy. Even FPVs of the DH pros clearing 40ft drops don’t communicate the ballsy epicness of what they’ve just done. Same with a squiggly GPS line on a map.

Salvo is our way of bringing tech and mountainbiking experiences together. We’re not looking to create a race specific device to shave seconds off a DH run. We want to create a device that gives people a frame of reference to appreciate the skill involved in that DH run…or any offroad run for that matter. Just think of F1 before telemetrics were displayed on TV and after. A blurry car vs a blurry car with 215mph next to it makes a big difference in your appreciation of what’s going on.

I guess it all boils down to recognition. Mountainbiking (or any extreme sport) represent some of the most impressive feats of skill and effort most of us will achieve, but they go largely unappreciated. Hell, as a rider you’re so focussed and in the zone that even you aren’t able to consciously appreciate what you’re doing, you only ever have a vague memory. Even just knowing how hard you stacked it would be nice!

So the plan is to develop a system that can monitor the various aspects of your ride, picking out the best bits. This can then be used for personal satisfaction (bragging rights), basic training and tuning advice (no point in getting too granular with it), as a cool video overlay if you have Go-Pro footage (we’ll be building an app for that), or online to gamify mountainbiking a bit (leaderboards, events, rewards, etc…).

In short – competing, socialising and appreciation feels good. So we’re the building on the solo nature of mountainbiking to have some of that.

Having said all that, I’m all ears when it comes to what you all think!

Just realised that I haven’t shared the project website :/

http://www.salvo.io

Anyone wanting to get in touch for any reason, ping me an email – simon<(at)>salvo.io

Some more replies!

Steve77

I’d like something I could tape to my fork leg or rear triangle and then it would tell me how I should set damping/rebound etc

If we manage to pull off what i’ve mentioned obove, then maybe we could do a “Pro” version that has many sensors for accurate suspension setup and training analysis. But those kinds of systems would be quite Pro focussed and they already exist in-house at FOX and SRAM. Also these guys – http://www.shockwiz.com/ & http://sussmybike.com/.

IvanDobski

It’s not quite the same thing really but the Powerpod power meter includes most of the sensors you’re talking about

Thanks for the info, I’ll be honest, I didn’t know about Powerpod till you mentioned it. It’s a pretty cool piece of kit, I might ping them an email to see if they’re have a dev-kit going spare.

dragon

You need to define it in terms of actual numbers, good luck

Cheers, we’ll need it! But if it were easy then there wouldn’t be much point in doing it 😀

although the typical running / cycling ones aren’t the best, there are more accurate ones out there.

We are planning on incorporating GPS into the device at a later stage when the tech become more reliable and less power hungry. There are some interesting mapping possibilities there….just have to beat Garmin to it!

Wzzzz

Here you go: http://www.lancaster.ac.uk/dsi/education/business-engagement/

Thanks for the link, we’ll definitely look into it. Although we’re not an official company or anything, we work on this in our spare time and on weekends, so i’m not sure the Uni would trust us with an intern, lol.

do a Kickstarter

Absolutely, then we’ll have the £££’s to be able to get interns!

HoratioHufnagel

you need to start pulling lots and lots of data out of it with preferably with a very skilled rider, and some “less skilled”, to try and see if you can analyse the data to find patterns. This is probably where you need someone with statistics knowledge to come in and help. These patterns will be useful in designing the MTB metrics.

Couldn’t agree more! We’re building up our database over the course of this year by which time we’ll hopefully be working with someone who can help us sort through the data and find those illusive signals.

I work with various computer vision algorithms, and I have wondered about using something like SFM (Structure from Motion) or SLAM (simultaneous localisation and mapping) to try and improve GPS accuracy for MTB

SLAM is actually something we’d want to do once we’ve got the metrics dialled. The idea of a highly detailed 3D trail map with ride points/scores marked along it is very interesting. But like you say, hard to do. That Microsoft like is pretty cool, maybe MTB SLAM is not so far out of reach….

fascinating – wonder if you had seen this approach for ‘sliding’ instrumentation.. (and capturing the outputs of good technique down at the ski<>boot interface)..

https://www.kickstarter.com/projects/333155164/carv-the-worlds-first-wearable-that-helps-you-ski

the data overlay for GoPro vids is an interesting angle as well (I think).

Sorry I don’t have anything more constructive (professionally) to add..

SLAM is actually something we’d want to do once …detailed 3D trail map…

FYI SoTA here is about 15W-20W to generate 3D on the fly (depending on sensors).

Possible on a bike, yes, but only for a couple hours at a time. E.g. a big 15/17″ laptop battery is 99Whr (as that’s the limit on a plane) so that’d be 4 hours with a reasonably large battery. Think 2 hours for a “light plus battery” bulk setup.*

I’d expect that 15-20 to come down to 10-15 in about 2 years if I had my betting hat on 😉

*though you could build a 6dof tracking setup fairly “easily”** to go on a test mule bike to get decent ground truth for correlating with IMU data.

**well, i know how I’d start…

Oh and indoor navigation based on IMU data is something I’ve done in the past…but the IMU alone isn’t enough. You need something else too (i used a nav graph of the building). IMU plus trail model – now you’re talking. But getting the trail model from the IMU? I’d go with “nope”. Just too much noise and drift etc. on them. You’d need another sensor and I don’t think GPS* is it.

*”good” gps aint that good, and needs a base station nearby.

Actually *ponders*….

…if you had a shit hot dynamics model of a rider, you could probably infer turn direction from the IMU, combined with speed from a wheel sensor would probably give you turns (the way a bike turns depends on speed), so maybe, just maybe you could get the trail model you need, combined with enough GPS data to sample to get the noise down (think kalman filter with predictions of corner-corner trail distances based on IMU+wheel combined with corrections from noisy GPS).

Problem is that riders motion models will vary a lot, and between bikes too. You’d need to do some calibration I think, until you had enough data to learn/extract the models, which would need LOTS.

Although thinking some more…. for open trails with a visible horizon…I’ll just leave this here… 😉

http://groups.csail.mit.edu/robotics-center/public_papers/Barry15.pdf

[video]https://www.youtube.com/watch?v=_qah8oIzCwk[/video]

Bonchance

wonder if you had seen this approach for ‘sliding’ instrumentation

Yeah I know Grant from Carv, he’s managed to get a bloody good product out in a pretty short space of time. It’s really good to see just how well recieved his device is considering the number of similarities between it and Salvo 🙂

AI

Actually *ponders*….

Wow, ok, you seem like you know your stuff when it comes to mapping in 3D!

SLAM is an eventual (inevitable?) outcome of salvo, but we first need to develop the more basic metrics product to secure the funds to go on and do the stuff like SLAM…..unless we can find a way to hack SLAM on the cheap!
If you have some spare time would you be interested in a Skype / Google / Pub chat?

BTW, That MIT video is both inspiring and scary at the same time.

Here is the questionnaire I mentioned:

https://salvo2.typeform.com/to/c6yTuV

I’d really appreciate replies as it will help us understand what to prioritise for our first prototype.

I understand the idea of it a lot more now.

Using wheels to measure distance is fine, provided your wheels aren’t skidding or are on the ground. But through air time, that distance goes, likewise scrubbing the tyres.

Maybe if you really wanted distance, possibly thinking of ground tracking in the similar way that the new dji phantom drone has a basic camera that points down at the ground which it uses to stay in a fixed point.

You don’t need an hd camera per say, just a means to track the ground passing.

Playing devils advocate, isn’t the ultimate metric for how successfully you have navigated your bike over a segment of terrain is the time taken to do it? For which there is already a fairly successful product.
Or put it another way someone bragging on social media that they have got the record for totally nailing a berm when they are 999th/1000 for the particular strava segment won’t really work.
My gut feeling is that to sell it, you need to sell the dream that buying the product will make you faster, rather than the social competition side, as that market has players with far bigger resources.

fisha
A speed sensor is something we defo want to add into the package down the line, giving us a lot more options with the kind of data we could show you.

Airtime, skids and the likes are actually not that big a problem. You can easily distinguish them from normal with some common sense coding as part of the algorithm. For example: If I grab a handful of brake lever at 15mph and skid, the sensor would see me doing 15mph, stopping insanely abruptly (we’re talking a good 68Gs here), then near instantly accelerating back to 12mph (pulling more insane Gs). If the algorithm knows the limits of what is possible for a rider, it will see those crazy readings and infer that it was a skid and not your actual speed. It could also know that your speed during the skid was a linear decrease from 15mph to 12mph over the X seconds you were skidding for, so you could indirectly track skidding speed.

IanMunro
Devils advocates are what we need right now >:)

I get that there are some pretty heavy hitters in the Social sports industry, but we’re not competing on their playing field. Big companies like FB, Garmin or even GoPro are looking for mass market products that can gain massive traction in a short period of time. Mountainbiking is far too niche for them to devote resources to it. If they were interested in mountainbiking, they’re more likely to simply wait until we’ve done the hard work and developed Salvo before simply acquiring us, that’s how big players “innovate” now.

With regards to time metrics vs ability metrics. Time is good if you’re in it for the efficiency or training, and there are plenty of good options for that. But we’re building Salvo for those times when you’d rather get some air off of that double rather than pumping it for speed, lowering your speed score but upping your “fun” score. I’m not saying that speed isn’t fun, it is, but sometimes you want to push your flair rather than push your speed. Salvo will cater to both styles, but we’re not trying to make a race specific training device.