Reblogged: An All-Automatic CVT Based Bicycle

      Filed under: Articles   

The following post was written by Ron George as a response to my NuFixie Challenge. He posted it on his own blog at on June 4, 2008. I’m archiving Ron’s full post, including comments, for posterity.

Personally, I’m not much interested in wanting to have an all-auto bike. Maybe its nice to have one in my collection, if such a contraption ever exists, just for the kicks. But the very notion of riderless shifting is sort of repelling to me. But all people aren’t like me, so its okay. Its nice to be sensitive to everyone when you’re out to design something.

A month or two back, I had written a post called “New Ideas for Cycling Products”. It was mostly science fiction, which I admitted. It was fun though.

Well it seems like I missed out on the All-Automatic Bicycle, a bike that thinks and shifts on its own without the rider’s mechanical involvement at the handlebars.

This daydreaming of mine is based on a post on Ari’s blog, pointed out to readers by James’ Bicycle Design. Ari calls it “The NuFixie challenge : Can you build a ‘fixed effort’ bicycle?

I want to take a small dab at this. Nothing is too technical or specific. I’m beating around the bush but I hope you see where it is that I’m beating. Drop in comments after you read.

Okay, lets take this step by step, one at a time.


Continuous variable transmission (CVT) is old old old stuff, supposedly originating with the genius Da Vinci himself. The first commercial patent was granted in the late 1800’s or so. The CVT concept has never found favor with bikes nor cars but AUDI has had some luck with their multitronic CVT car. Remember that one?

Fallbrook’s design is fresh. It is a step less TRACTION or FRICTION based transmission using rolling spheres and some kind of mystery fluid. Take a look.


Basically, what Ari tries to ponder is this, in his own words :

The idea is to build an automatic continuous transmission for a bicycle, by wiring up a controller for a NuVinci CVT designed to maintain a constant level of effort from the rider.

Or in other words (if I have interpreted it correctly), take a CVT step less transmission equipped bike, completely eliminate the need for the handlebar shifting operation, and take care of all the upshifting and downshifting based on effort. So the input is effort, the output is a transmission change to appropriate and suitable gearing.

This can be suitable for any type of rider, also to people who can’t shift because of a wrist problem, people in cold places who freeze or anyone who has handlebar phobia.


The design goal here is to shift automatically based on EFFORT. But it depends on how to look at effort. What is effort?

Everyone has a different heart rate for a given pedalling load and the perceived exertion is also different, based on what your mindset is, how motivated you are, what your pain tolerance is etc.

Everyone has their own preferred cadence given the same kind of terrain. Furthermore, most people who ride a lot can do so in the same gear even if the terrain gradient changes gradually. Other people can’t, won’t be able to maintain the previous cadence, and have to shift.

But HR and pedaling cadence are not necessarily objective pointers of effort.

Power output is, and it stays constant over wide variety of conditions.

I think a solution here is to monitor a combination of power input, heart rate and cadence using the sensors available in the cycling market today.

Body temperature and gradient change may also be potential inputs. As we put more effort, our core temperature increases.

As far as gradient is concerned, we all know that when we approach a downhill, we always manually upshift in order to not spin out of gear. Question then is, can you replicate this same behavior in a computer such that a considerable altitude change alters the gearing correspondingly.

If you want to make the system more fancy, go ahead and monitor wind drag and vibration frequency corresponding to a type of terrain. For example, the best and fastest way to move over dirt roads and pave is to get into a big gear and grind at it.


Right off the bat, this is a control issue.

Say this theoretical bicycle had a computer that tells the hub when to shift. I mean, it needs to have a nerve center right? You aren’t doing the work of shifting, you want a type of robot, which is the goal.

Roughly, imagine the computer is a rectangular black box that is able to monitor power, HR, cadence, speed and what not. Additionally, a user may be able to change his preferred settings if they want to ride at a particular cadence that day. Whatever…

The complex part of the project is to program this computer. Perhaps you could use a PLC, something small enough in the market today, write a ladder diagram using software and fit all this into a handlebar mounted computer. PLC’s are commonly used in CNC machines, AGV’s, traffic light control, wherever automatic process control is the name of the game.

The computer will then be coupled electronically to a mechanical shift transducer, in this case, something that will move the idler laterally that in turn will tilt the axis of the spinning spheres in the hub. Digital signals to mechanical motion. Yada yada…


This will be the main headache.

The big issue here is that human physiology varies a lot from person to person. Everyone has their own abilities, their own wattage vs. HR graph,their own lactate thresholds and power profiles and their own preferred cadence for a given situation, which may or may not be always optimum.

If you did write such a program to ready the computer for automatic shifting, could the same system be used for another person, given the above physiological differences?

I don’t know the answer to that. Some like to spin fast on a hill, some like the big gears. Some are very efficient at climbing heart rate wise, while others will struggle and possibly die. Some riders wouldn’t shift at all were it for a 1 mile commute to the grocery store. Others may ride 25 miles with plenty of climbing every day and plenty of shifting is called for here.

Like I echoed before, such a system will either have to be custom built or the computer must be designed in such a way as to take in a set of inputs based on different rider’s abilities.


So the computer operates a mechanical actuation device that in turn rotates the idler.

You could go hydraulic. Okay, no…thats messy. Too much fluid.

Find a compact, geared DC motor that is reversible. Hook a self locking gear mechanism such as a worm gear that mates with a corresponding sprocket at the end of the idler. The selection of the motor mainly revolves around desired RPM, torque, and power output which will need to be calculated. Cost of the motor depends on these specifications.

I mean, there are tons of ways to do this. I just told you one.


Very important with any project.


I don’t know the answer to that. Who am I to sit here and say this will work, this won’t work? But as with any project, things to keep in mind are :

  1. Customer - Who is it, what do they want? Are they racers, recreational cyclists, beach goers, commuters, who??
  2. Minimum Specifications and Additional Features - Less moving parts, so much weight, so much dimensions, so much tolerances, durability. Is it too heavy, is it too cumbersome, is it too big that it will induce aero drag?
  3. Cost - How much can you make it for? How much will a customer pay for it if you solve their problem?
  4. Competitors - What exists already in the market and how will your design address their disadvantages and advantages. How efficient is your design?


This project is fun but difficult. Not impossible. But its complicated and may bring in more problems that that you’re trying to solve. One of the problems is there being a lot of components, wires, and moving parts. If you have all this setup and hooked to your bike and then if it suddenly takes a spill, you may be in to cover more costs in damage.

I would not even bother doing this, really. I mean, Shimano’s next Dura Ace is an electronic model that just relies on pressing buttons for shifting. We’ve gotten things down to this level of simplicity. Next year, and the year after that, when these electronic gizmos are mainstream, they might come out with something for hybrids and street bikes. But I don’t see any drastic improvements by taking even this ‘button’ away and making an all-automatic “thinking” bike.

Bicycles are meant to be human powered. Designs are often simpler and cost effective that way.


I haven’t tried it. But I like it and would like to get one, perhaps a fully fitted bicycle or just a hub to play around with in my spare time.

I also do not agree with the snobs online who put this down without knowing what it is. Gentlemen, this does not work like those cheap Auto Shift bicycles. There, the primary mode of operation is the moving of the dérailleur through centrifugal force based on pedaling cadence.

Auto shift bicycles fail precisely there - the possibility of shifting at a too low RPM when you don’t want it to happen. And for maintenance purposes, what do you do if you want the chain on a particular cog? Sit there and rotate the crank at 90 rpm manually so it can shift??

Not that good design cannot come around all those issues but still…

This is a CVT hub and not a hub based gear cluster. It is meant for step less, noiseless, and smooth transmission giving you an infinite amount of gearing. There is no click-click-click.

I don’t care about weight. My main concern is about the moving parts. How reliable is it in high torque, low speed situations such as standing starts and hill climbing? Will it work if I get out of the saddle and stomp on the pedals? Will that kind of friction hold? Will I be wasting my power?

These kind of questions are addressed well in the traction fluid section on Fallbrook’s website. The company also got German SRM and an ex-Tour de France rider to test a Nuvinci equipped bike, as shown in this Youtube video. Based on seeing it, it seems amazingly smooth.

And what about the hub itself? If I get out into the elements, can water penetrate the outer compartment and mess with the lubricant? Can the liquid leak? Is it thermodynamically stable? Is it easy to service, maintain etc etc etc, you know where I’m going with this.

I wish that NuVinci would publish more numbers to go with their website. How efficient is this system, say compared to modern chain and derailleur systems?

Chain and derailleur systems have very high efficiencies, typically above 90%. See Chester Kyle’s test study here (PDF).

I think somewhere Kyle mentions that for every 1% decrease in efficiency, 12 seconds are lost in a 25 mile TT. Well, the NuVinci may not be for racing at the moment anyway…

NuVinci CVT is very practical for hybrids, city bikes, ladies’s bikes, even electric bikes.

But for the hardcore racers who like to go fast and ride ridiculously light bikes, there’s no point in sitting and complaining about CVT - if you don’t like it, live and let live. Its not for you. Simple.

Now I really have to chill. :)



TD_CBCS commented: "I saw the same design on yesterday. That guy only has the drawing though. He calls it NuFixie. I'm not sure if this is the same people or not."

Ari commented: "Hi Ron, Thanks for your analysis and thoughts on my NuFixie idea. I agree with you that it isn't exactly practical, but believe that it is an interesting thought experiment, nonetheless. While heart rate, power output, and cadence each can indicate effort, but do you think that one would necessarily need all three? What would be the minimal sensor input required to construct a NuFixie? As for the drive train controller, my thought was that it could probably be located directly next to the NuVinci hub (with a single dial on the controller allowing each rider to specify their preferred effort level.) Great post! Nice analysis."

Surestick commented: "I think the idea of shifting a CVT based on power output is overly complicated. I feel that auto shifting a CVT could be done by having the shifting set to maintain a specific RPM and then letting the rider have control of the RPM with a manual shifter. If you think about it, most of the time when you are shifting it is to change gears to maintain a specific RPM due to changing wind, hills that can be climbed sitting, fatigue, etc. It's only when standing to sprint/climb or going from road to off road conditions that we really shift to change RPM. I've though a lot about CVTs, internal gear hubs and bicycle drive trains myself. If anything I don't really see the need for a CVT on a bicycle, one tooth jumps in a cassette give close enough gear ratios unless one is using a huge chainring. For most cyclist I would guess that being able to shift under full power would of greater use, especially for mountain biking. The armchair engineer in me has always imagined something like a cog split into segments of several teeth sliding in and out on tracks radiating out from the hub allowing the cog to change diameter as being the best candidate for a lightweight, efficient replacement for the dérailleur."

Ron commented: "Ari : For simplicity's sake, cadence may be used for it is not a good objective measure of effort, which is the main thing to determine here. And for using something as trivial as cadence, a project such as this all automatic bike is not even justified. Just use a cyclocomputer and manually shift. The big engineering here to first to define what is effort, and closely monitor a bunch of critical measurements, one is power output, the other is HR. Most of the engineering will be consumed here and in programming the system. Also for simplicity's sake, say that we don't need any of the aforementioned variables. Say that we just measured the slope of the ground using something like a digital inclinometer hooked up to a computer. Every time the gradient changes and increases or decreases beyond a threshold value, shift. For a pilot project, as in developing a simple prototype, it's not bad to start out with something simple like that and then branch out to other areas once you get the expertise. Overall, this is a hugely complicated idea :)"

Ron commented: "Coach Joe Friel talks about accurate measure of cycling intensity and effort here. Also, check out the YANKEE BIKE design, another non-CVT auto shift bike. Popular magazine wrote an article about it in the 1970's. Pretty interesting. I wonder if it ever worked reliably."

Ron commented: "SURESTICK : I too am a little concerned about a transmission that works by friction or traction. However, there's no harm in getting one to try out. I have never had a CVT, so I'm not going to talk for or against it."

Ron commented: "Here's some more discrete geared automatic transmission systems that already out there in the market : Browning Automatic Bicycle Transmission TREK Lime 3 Speed Automatic Text to be displayed Automatic Bicycle Gear Selector Mechanical Design of an Automatic Bike Transmission Shimano NEXUS Auto 3"

Bike_Boy commented: "An idea for a computer : Just buy an existing Cinqo or iBike Pro Powermeter and use the microprocessor in it for your needs, as in HACK IT. I unfortunately dont know how to hack things. :( Check out the Shimano Di2 (Integrated Intelligence) Nexus groupset that features adaptive suspension. I think nothing can get better than that. The system shifts gears when going uphill and downhill. I think you have to buy a whole new bike with this setup, not really sure. I also think its only available in Europe. Here is Andre Jute's Touring Bike, all tricked out with such the smover. This is bike porn, WARNING."

Ron commented: "You also need some kind of servomechanism. A reader pointed me out to this page, where you can get some handy hobby servo controllers."

Josh Mitchell commented: "Ron, Are you familiar with a concept called Maximum Power Point Tracking (MPPT). It's a term used by the smarter Photovoltaic controllers to eke every last bit of energy out of PV panels given the current solar conditions. What it, essentially, does is play with the amp demand it's taking from the PV panels and plays with it to find the "best" point. The reason it needs to do this is that the Volt / Amp characteristics of a given wattage PV panels is not linear. Anyhow, it seems to me that the key to making a NuFixie work is employing a logic that would be similar, but let's call it MSPT (Maximum Speed Point Tracking). Basically, the assumption is that most riders would prefer to go as fast as possible given a particular output they are willing to do at a given time. The idea could be that you don't use setpoint logic, you use maximization logic. Based on an rpm sensor in the rear wheel, the controller can play with the gear ratios in the CVT to see what the reaction from the cyclist is. If the cyclist doesn't "like" the new power output demanded to go at the new higher speed that the computer is shooting for, power output is going to drop off suddenly. Now, the beauty is that you don't have to measure this power directly because when power drops there's going to be an reduction in speed / rpm... Now realize, using sensors other than the rpm will speed the potential reaction time of this system. (as if power drops off you will have some coast over speed that needs to go away). So maybe drive side chain tension (some strain sensor in the drive cog of the hub?) drop off could be used as an indicator that power dropped suddenly. Of course one could always put power sensors in the cranks... but that's pricy. Point is, you don't need cadence sensors and HR monitors because this would allow the person to naturally find their current "best" cadence given their desires for that ride / their physical condition that day. And, ultimately, the only sensor you need is an rpm sensor on the rearwheel. Now, it'd take someone smarter than me to actually make it work, but it could work. Ultimately, all you "need" is an rpm sensor in the back wheel"

Ron commented: "Josh, I get the basic picture of what you're saying. You're spelling out an optimization strategy through illustration of MPPT, although I dont have the serious electrical background needed in photovoltaics Nevertheless, I think the basic idea behind MPPT could be extended to other systems as well. Your logic employs the PERCEIVED idea of effort. It was the most commonly used until HR and PM's hit the market and dissolved its accuracy. Turns out, for most people its absolutely fine. Do I have a PM, or ride with an HR monitor, NOO! There are chain based power sensors available but they are in the mid-range expense (>500 dollars)and involves complicated, expert installation. Like someone else echoed here, the iBike power meter just approximates the opposing force to motion, mutiplies that with speed to give power. Doesn't work always but its okay. So thats an avenue to go if you can't afford an expensive one. I like the way you described the way computer looks for your optimum mood for the day. Ofcourse, for the logic to work the assumption should stand. Hence, comes in the "SCAN" mode, a 5 minute interval during which the system interprets your desired effort and gets into the right gear. Now logic could go wrong when, suppose the rider goes downhill for an extensive period of time or bends down to get a bottle of water, its likely that he'll slow or completely stop pedaling (coasting action). For going downhill, obviously you want the system to upshift and for the latter, you don't want the system to change. But fear not, there's always the option of MANUAL control. Or the computer could have a "DOWNHILL" chice, where a different pathway to control is taken. We can program the downhill logic, based on changes in a calibrated tilt-o-meter (ha!). Once the button is depressed and the option removed, the system could revert back to previous "FLAT" mode. As for employing other sensor inputs to better reaction time, the next simplest is HR zone monitoring. To get this to work, user will perhaps have to calibrate his zones by going for a short time trial or something. For a beater bicycle, this seems stupid but the zones don't have to be dead accurate. A ball park is sufficient. Once you have that done, everytime you begin to start riding your bicycle, you choose an effort level for 0-10 on the computer menu, each effort level or combination of effort levels corresponding to a range in HR. Voila.. now it reads your HR, trying to keep it in the respect range while you ride, all monitored with the cadence sensor on the rear wheel. As far as cadence is concerned, I think we can make the system more fancy by allowing it to recognize the user's preferred cadence over a long period of time or miles. This way, the system can then "LEARN" to operate in this efficient range. And if there is ever a problem of all this being messed up by a different rider sitting on your bike, bring in USER PROFILES, each profile can be saved and pulled up according to the rider. I think Ari would be happy with your suggestions, which is what I told him as well. For simplicity's sake, start out with a RPM based prototype. Whether he plans to actually make one is totally another matter, which I know nothing of so far. I didn't go too much into the programming logic, and I'm sure there could be other suitable optimization algorithms. These need to be brainstormed. The internet is not the best way for proper communication."

Fordi commented: "As far as I can tell, there are two basic advantages to a CVT over a conventional derailer: 1) You'll never, ever, pop your chain out of place 2) You can shift at 0 RPM."

Fordi commented: "Actually, this can be pretty simple. Inputs: Input torque Current speed Shifter (desired input torque <-> acceleration ratio) Mercury (detects bike's current pitch, sampled at 8kHz and averaged over past second) Output: torque No need to adjust for physiologicals; chances are that any individual need only move the shifter once to find a comfortable setting. As for the other variables, I suggest a game loop that intelligently correlates output torque to measured acceleration; let empiricals stand in for the combined measurements."

LoneStarNot commented: "Interesting! As Fordi said, some simplifications seem possible. On the surface, torque seems the only critical input. Cadence is a red herring; riders negotiate cadence (with the aggregated load) based on torque. Riders also naturally integrate other loads: stickier surfaces, higher winds, road gradients, etc ... by applying more or less torque. Torque's the biggest thing. Timing, however, makes simple torque less nice: a) riders rested from a stop start with an initial torque higher than their steady-state torque. That implies a need for a compensatory algorithm for "starts"; b) even idealized torque is cyclical, related to crank angle -- so the algo would need crank angle compensations; c) rider effort is surely not steady throughout a ride, but declines after some early peak. Some form of dynamic optimization (hunting) seems appropriate, but again, timing issues would have to be worked out: How fast can gears be changed? How fast should the system adjust to torque changes to be "responsive"? ... without oscillating? ... net ... I expect an "torque profile" preloaded into the system... - startup curve (rider input) - trip curve (rider input) - Instantaneous torque = avg torque * f(crankangle) ... would yield a good response, based on only torque and crank angle. Also ... Accurate Power measurement implies accurate torque AND cadence measurement. Just measuring torque could thus reduce the error too. I'm guessing chain-deflection torque meters are the simplest/cheapest and would be sufficiently accurate."

Josh Mitchell commented: "LNS - If chain deflection torque meters are "close enough" wouldn't something that counted the linear chain speed be "close enough" as far as cadence is concerned (presuming you entered the size of your chain ring)."

LoneStarNot commented: "Josh, I'm guessing ... guessing that torque correlates to perception of effort so well that nothing else is needed. In doing so, I'm eliding LOTS of complexities -- rolling resistance, wind drag, incline, crank angle, etc. -- on the guess that "perception" integrates all those things to get "effort". Thus there seems no point in adding complexity to re-do that job. While I do expect that the same torque would be perceived as more effort when demanded on an incline, I also expect riders would perceive that as "natural" and appropriate. I mentioned chain deflection meters because I'm guessing they're cheaper than other torque measurement technologies. Chain speed would likely be more accurate than torque, as a percentage -- but I expect it correlates less well to perceived effort. So I'd guess chain-speed unnecessary in creating a "constant effort" machine."

bike commented: "Thanks mate.The video is so cool!"

Anonymous commented: "Interesting article. I was thinking about taking one of the cvt's from the little pocket bike motorcycles and hooking it up to a bike. I would remove its centrifical clutch and wire up the gearing similar to a grip shift. Has anyone else thought of this? Would the efficiency of the belt drive be so bad as to make it not worthwhile? any comments or suggestions?"

Bruce McL commented: This is an interesting thought experiment. For computing power, inclination, and even rough guesstimate of speed, you could use an Android smartphone. Another computing alternative would be an Arduino. I agree that measuring crank torque would be just about all of the input you'd need from the cyclist. But I can't think of an easy way to do that. Sensors in the pedals? Then you need wireless to get the info to the computer..." commented: "with the time you spent writing this article, you could have taken the Nuvinci automatic AND manual system for a test ride and found out for yourself how much easier it keeps you on top of your cadence."

Bryan commented: "Ron, The multiple-input, microprocessor-laden approach is overly complex. This is particularly so, given that a time-tested system exists from aviation: the "constant speed propellor governor." Basically, the system involes a spring, acting in one driection, and spinning "flyweights," acting in the other. The spring and flyweights are at equilibrium at a given RPM; if the equilibrium is broken, a piston is displaced, flowing hydrualic oil in one direction or another to change propellor pitch and restore equilibrium. The pilot controls for RPM by adjusting the "preload" on the spring, resulting in a totally "set-n-forget" system. Something very similar could be rigged for bicycle cadance--no silicon, no microprocessors, etc (though perhaps not using hyd. fluid). The rider could "dial in" a desired cadance via preload, and the mechanical system keeps it on cadance until the rider chooses another cadance. (The only improvement would be a "squat switch" in the seat to automatically reduce cadance when out-of-saddle.)

Julian commented: "Like bryan above says, it's way simpler if transmission is always shifting itself to allow a particular pedaling cadence (for example 75 rpms). All it needs to measure is the speed (once the wheel size and the sizes of external gears are known). Next simplest idea is, just use two factors. Measure speed and measure the force on the pedals. You can assume the rider wants a relaxed pace if he is putting about 30 pounds of force on the pedal during each down stroke. Then assume the rider wants a fast pace if he is putting about 70 pounds of force on the pedal during the down stroke. For relaxed riding, set the pedal cadence to 60 RPM, and for fast ridinig set the pedal cadence to 90 RPM. (in-between force should get you in-between cadence). This is because if you want to pedal hard and go fast, a higher cadence is ideal, whereas if you want your ride to be really easy and relaxed a low cadence is ideal. The assumptions need to be adjusted for different riders (a 100lbs rider puts less force on pedals than a 200lbs rider, and some people like to pedal 60-90RPM while others like to pedal 55-70RPM) "

Yesurbius commented: "Your idea would have merit in one other place. If you look at the KTRAK - its a Ski and Track conversion kit for bicycles allowing them to go on snow and ice. Having a self-contained CVT would be beneficial in such situation (assuming it doesn't freeze) since the resistance and grade is in constant flux while pedaling. I don't think that a computerized feedback and regulation approach would work. Ideally you'd want it to be a mechanical regulation where the user sets the desired pedal resistance via a shifter, and the bicycle is pedaled at a constant rate. The output speed is automatically based on the resistance setting with reductions made for resistance and grade (torque requirements) "