How It Balances on ONE Omni-Wheel

[@Peavey_]

I love that the side of the final construction at the end of the video looks like the face of a person who can’t stay balanced. Love this project.

[@doxielain2231]

Ok, now I can't unsee that jester

[@redroyal4287]

I can't unsee that now

[@xrdbnk]

@@doxielain2231 i’m so glad i wasn’t the only one to specifically see a jester

[@timjeffries9833]

Yes! I saw it immediately and I’m happy to not be alone.

[@michaelcooley3381]

Bf

[@ogbushwookie5003]

Given the encoder could you use that to determine the position of the large wheel on the bottom and use a velocity multiplier based on the known geometry for the tapered wheels? May not be perfect but better than unknown.

[@matt_nolan11]

Seems pretty doable! If he's using a relative encoder then it will be difficult because he won't actually know the rotational position of the wheel, just the displacement from the initial position at power-on. Using an absolute encoder or being careful to power the machine on with the wheel in the same position each time (or rather, one of several equivalent positions because the wheel has some rotational symmetry) would fix this though. Maybe we'll see this in v3?

[@ZoidsNut]

That's a cool idea and sounds like it should theoretically work on a level surface. But the moment you encounter any sort of incline you'll be in trouble because the part of the wheel that's in contact with the ground isn't at the direct bottom of the wheel.

[@MechaNexus]

yeah, this was the first thing that came to mind once he mentioned having problems with the taper on the small wheels. It's pretty doable and I bet James already thought of a similar solution.

[@wiggwigg2010]

I was literally thinking the same. The encoder would need to be quite high resolution though as even a single degree of precision would throw it all off. This could be get quite expensive so the trade off between this solution or just a bumpier ride via constant diameter "wheels" would be interesting. Mind, he's probably got enough encoders lying about that one will work.

[@sherwoac]

I'm not a control theorist but..how about, instead of using a positional encoder you could see how much acceleration (from IMU) a certain movement of the wheels made, calculate the implied gearing (which side of the wheel is on the ground) and adjust the movement accordingly. I guess this would be akin to a double pendulum closed-loop? Success would depend on the sampling rate of the IMU, motor drivers, control loop, etc.

[@brianclough3924]

Can we all take a second to appreciate the depth of incredible projects James as made, explained and published for free? These are not simple projects, but exceedingly hard electromechanical designs that would typically take a team of 3 to create. James pumps them out weekly 8]

[@jamesbruton]

thanks!

[@theredknight3736]

@@jamesbruton No. Thank you!

[@marsgizmo]

pretty interesting design! 👏

[@ironarsh7202]

First reply BTW love your design marsgizmo and James

[@mopatop3921]

I have no idea how you manage to design, print, film and edit all these projects and get a video out every week, it's amazing!

[@Smurphenstein]

It moves a lot like my neighbour walking home from the pub, which is probably why he ends up crawling half the way. Very cool as usual James.

[@Skyentific]

Brilliant! I love this design. I still have no idea, how you able to make such high quality build and video EVERY week!!! Impressive.

[@themainc]

Always love watching your videos. There's something that almost borders on magical about how good you are with robots combined with how much you clearly enjoy working with them.

[@epiblitik05]

If you could predict the position of the shaft you could predict the cone of the smaller wheel and adjust your algorithm accordingly. Incredible video as always thank you for making these!

[@ILoveTinfoilHats]

Differential equations. Little bit of math would solve this whole thing

[@shable1436]

The coding needs to be tweaked so it's horizontal axis isn't as glitchy and mechanical, using no gradual braking or acceleration, just full power, full stops making it compensate itself by incremental building up reverse inertia via momentum.

[@jincyquones]

12:47 The design is good because that's exactly what my face would look like if I were a one-omniwheel balancing robot.

[@WetDoggo]

Next iteration: - using more and shorter wheels with constant diameter - miniaturizing the belt system to fit more belt systems - using one gear to drive 2 belts (effectively doubling the amount of usable belts)

[@reubenj77]

Would the idlers perform better if they were a bar with a concave curve to them? Maybe less slop?

[@WetDoggo]

@@reubenj77 either straight or convex, since the belt will walk itself off a concave roller

[@notahotshot]

@@WetDoggo, yeah, I saw a demonstration where a belt will center itself on a convex pully. He should be able to prevent the belts from walking, by making the mating surface convex.

[@WetDoggo]

@@notahotshot the idlers/tensioners can be convex, but the gears can't be convex for good meshing. At least not strongly convex... A little bit of "convexness" might actually aid a bit in keeping the belts in center. It might be worth a Test 🤔

[@notahotshot]

@@WetDoggo Steve Mould did a video about this. It's actually called a 'crowned pulley'. kzbin.info/www/bejne/in_Yq5xjn7KLpbs There's actually a clip in that video of a belt, twisted 90° between two pulleys, no idlers, staying on the pulleys because they are crowned.

[@aserta]

12:37 an appropriate face for this robot.

[@real-1982]

You are very smart on designing

[@gistnoesis9116]

12:34 the sheer terror of an always falling robot that just saw its robot graveyard.

[@TheRumpletiltskin]

3:12 or you could print your gears with angled teeth so they grip the belt at a proper angle.

[@Clasticon73i]

Just a quick tip: adding a bulge to the belt driven gears will keep the belt in the middle of the gear. Not sure if it will work with a twisted belt. Cheers

[@AgentJohnSteed]

Crowned like on a band saw

[@milandavid7223]

Whenever I encounter this concept, I always have to convince myself that the belt would actually move towards the bulge.

[@tracypanavia4634]

@@milandavid7223 It keeps train's central to the rails too

[@toastsandwich9928]

Videos keep getting better, awesome work!

[@gingermany6223]

well done on getting it to work. Also some bonus pareidolia at 12:38.

[@esser50k]

so much work goes into these revisions!

[@eliaamit2010]

wouldn't dust and debree be a major disadvantage of designs like this where the wheels that contact the ground act as part of the internal mechanism? I imagine it could jam up pretty quickly

[@aviphysics]

That's mostly true of any omni wheel.

[@ionbusman2086]

@@aviphysics not that motorcycle

[@LeDracodon]

@@ionbusman2086 It's also the case for the mini motorcycle. Drive it a little bit into the mud and even those nice molded plastic gear will eventually jam.

[@craoun6116]

12:40 It has a face! Looks like a drunk dude stumbling around 🤣

[@uddalaksarkar3280]

when james becomes a whats inside channel

[@AliSot2000]

13:38 it looks like it’s screaming with the motor being the eyes, the handle as nose and the cutout as mouth.

[@fractal_lynn]

Yessss this is such a cool project, glad to see an update so soon! You're so fast with this :D

[@ecicce6749]

Ever thought about using machine learning to balance these robots? seems like the perfect use case for ML. It could optimize for stability and least wobbility. Gather lots of Input and sensor output data using this crude stabilizer (add some randomness to it) and train a sensor output predictor. This is a physical model of the bot. Use it to train a second model that optimizes input to predicted sensor output for stability and least wobbility. The gathering and learning could be done more or less in realtime. Let it do its thing and collect more data for both models and train it at the same time.

[@muneeb-khan]

11:56 he looks so worried.

[@Will_Huff]

So very cool! Really loved this design! You are almost there!

[@WeAreQuestionDutch]

Very nice how you back engineered this solution after a search, learning and own try and error.

[@BRUXXUS]

Incredibly cool! The belt solution was really great!

[@eathonhowell7414]

The noise of this thing and the motion as it moves makes me think to "Red's Dream". It's a short by Pixar and it is adorable.

[@MrZigzter]

Impressed as always. Hall sensor the wheel position and add a oscillatory coefficient to compensate for the tapering rollers.

[@demiseman9586]

Hey, Motorsport engineer here. Really cool wheel mechanism on that robot. Although, I don't think the tapered wheels are hurting it as much as you think. I have a hunch that controlling the lateral motion with only one motor is creating a yawing moment due to gyroscopic effects from the motor. I think when the robot leans to left, the motor spins up to lean it back to the right. But in doing so, also yaws it to the left due to gyroscopic torque. There's probably some unintentional feedback that is created from there. The easiest fix would be to spin up the second motor as well to control lean angle and see if that does anything.

[@jacobellinger8027]

A true engineer, design. works as intended but 20 different screws just to replace even one part should it go bad lmao. I love your content and I'm only picking on you because I like you.

[@tellhimimeating9141]

i really enjoy how it's constant movement makes it look properly alive. It's like a little star wars droid.

[@ChannelJanis]

Can you add an encoder to get orientation of the big wheel? then you should be able to know the diameter of small wheels contacting ground. Then add it to calculation of required motor speed. A bit of an overkill, but I see the wheel will vibrate when moving forward.

[@ckdigitaltheqof6th210]

0:23 that main middle wheel, with those tiny side wheels needed to be rotated 45° and in the shape of a foot ball clipt at the edges for the accels. When the main wheel rotates, it will not have uneven wheel bumps, and the bottom main and rim wheels, will sync combo to all required ground direction.

[@TaytayteamTaytay]

i love when he says there's too many gears for his liking from the toy design and instead adds way more gears.

[@amarissimus29]

Wow that got complicated. And I thought it was stuffed before. I love how you never give up on an idea until you've exhausted a whole bunch of variations. Can't tell you how many ideas you've given me (read: designs I've stolen).

[@Morberis]

Great job with all the improvements!

[@The-Weekend-Warrior]

I'm not sure if using belts is really better than using gears. The non-constant shape of the small wheels though is definitely something to re-visit... but it looks and works amazing already!!!

[@CDRaff]

12:50 It looks like the face of a person about to fall down.

[@Srcsqwrn]

This is really cool! I would love to see a version of this where the wheel is a bit wider (Or maybe there are two center/doubled up wheels?), and the machine is rideable like a One Wheel. I'd just like to see a board like that cruise around somehow. Ty for the link! Wonderful work!

[@glennburrow4364]

If you added one more drive motor and made every other small wheel turned by the new motor, you could potentially gain another axis of control, i.e. yaw. It looks like you have 2 small wheels contacting the ground at any given time, driving them in opposite directions will give you that ability to turn. Though, you would almost certainly have to add an encoder and programming would be a right pain (every time the main drive wheel advances another small wheel, the rotation would have to flip between the small wheel motors to continue turning in the same yaw direction). Actually, that sounds really complicated. Maybe just a huge flywheel and use it like a spacecraft reaction wheel, haha. The Honda 'single wheel thing' essentially did the same, except the reaction wheel was the user's arms.

[@qwertylesh]

very awesome, huge step up from the last design. I'm wondering about something, is there no need to grease the cogs and other moving plastics? eitherway, looking forward to the next iteration! great work!!

[@R9000]

Yeah I thought maybe this could help with acceleration and give the PID a bit of a helping hand.

[@PeterWMeek]

With a position encoder on the big wheel, you could "know" what the circumference was of the part of one of the little wheels that was touching the ground at the moment. Use this circumference information to tune-on-the-fly the side-to-side balance feedback.

[@douglasmaclean5836]

Sir James, always amazing projects & videos...your can-do attitude is addictive... & your efforts are undoubtedly inspiring so many ... Cheers ! from across the pond in Ottawa.. :)

[@h7opolo]

2:57 did you consider making hourglass-shaped gears to encourage the band to stay in the narrower middle section and away from the sides? i think that would be the best solution.

[@SagaciousTch]

An hourglass shape with a narrow middle will actually cause the band to migrate to the sides, the opposite of that shape makes the band stay in the middle.

[@thesfreader3068]

@@SagaciousTch Yep, that's what I understand too (counter intuitive, but still). he would need to "crown" the band.

[@h7opolo]

@@SagaciousTch well he hearted your comment and another commenter supported your claim, but i would like a demonstration. do you know of any?

[@SagaciousTch]

@@h7opolo I do, actually! Steve Mould made a video about it 9 months ago.

[@h7opolo]

@@SagaciousTch thanks, i remember watching it, but couldnt remember who uploaded it. i'll rewatch for a refresher. thanks. edit: perfect. kzbin.info/www/bejne/in_Yq5xjn7KLpbs it is quite counter-intuitive. OK, i stand corrected. the effect is also a function of band width, i.e., thinner bands exhibit less effect than wider ones.

[@ObserveTheBanana]

Once you see the face in side view, you can't unsee it. Amazing work, thanks for sharing.

[@LeeAtkinson98]

Man james loves his belt drives

[@milandavid7223]

I even enjoyed the sponsor segment, cool stuff!

[@FrankGraffagnino]

cool video. i was super impressed with that mechanism in that toy!!! kudos to those designers.

[@TheAshlec]

You could reprint the soft wheel section with the opposite tapper to cancel the taper out as easy way to check the stability

[@zeekjones1]

If you have issue with the taper, have it move to the center of the small wheels before side to side, when not moving. Would need to index at which rotation each center is, so it can move to the nearest. This will cause it to move forward or back a little, but then the side to side would be much more stable.

[@david-oh7qg]

A few years ago Self balancing wheels were mindblowing and today we got People Building them at Home. We slowly get to the Future

[@Dindonmasker]

That was really cool! Well done!

[@garrettstater7700]

Been waiting for this for a while awesome work mate !! 😛

[@aduh740]

I was looking at the final product. The space at the sides of the robot is notably open. Maybe place some kind of balancing mechanism at the sides of the tall skinny bot. Reason is that a person on a unicycle find their arms instrumental for balance. Some kind of gyros there maybe? With small rubber tube bumpers wrapped around the whole of the sides to protect bot if it falls.

[@rossbrigolimusic]

I think you can compensate the tapered wheels problem in the software. You need sensor like an accelerometer to tell the position of the bug wheel or which part of the tapered wheels is touching the ground.

[@FilterYT]

What a remarkable person you are!

[@richardduncan9740]

You are a mad man James! The fascinating kind!

[@Sqwince23]

I was wondering if it were possible to have two central hub gears like the toy, but put an extreme helical gear on the outside perimeter of that gear that engages the outer rollers directly. I was looking at how your model comes together an the gear portion would be similar to the notch you have now for the belt, but would be small semi-circular cutouts where the outer rollers touch the inner two bevel/helical gears. if you get the angle just right the inner hubs could drive the outer wheels directly.

[@DingusGoesWest]

Ooh I like that!

[@petermines3575]

Awesome. Your getting much better.

[@tanjiro3285]

awesome man🔥🔥🔥🔥🔥, I am such a beginner that I feel bad as I cannot fully appreciate ur work

[@Mike-bs5pi]

Neat! Maybe a counterweight above center of gravity would help? Just a small weighted lever on a servo to help it catch its balance quicker

[@alexlutz2221]

One way to solve the tapered issue without requiring too many major changes would be to put an encoder on the main drive motor or wheel and therefore you could calculate where on each taper it's driving on so (assuming it's on level ground) it would be able to compensate at least to some degree for the difference. I'm not sure how you would be able to compensate for it if it's not on level ground however as with only one wheel there's no way to differentiate between the level of the ground and the level of the robot (short of possibly compensating for that by calculating the amount of bias in one direction it is applying in order to stay stationary and using that and a bunch of testing to figure out approximately how many degrees the ground is off of level) although I'm not sure how well that would work.

[@teron281]

Wouldnt the bolts of the "planet gears" eventually eat into the tension ring? (Especially on the side where the screw thread touches the plastic 5:56)

[@tylisirn]

The bolts are not spinning. They're just static axles.

[@quinnfoster4671]

Related to these general multidirectional self balancing stuff. I am interested in something ridable so 3/4 suspension wheels to proxy a single contact point was the idea I had. Hand remote to recorrect rotation, or roll at low speed maybe.

[@bpark10001]

Add weight up higher to raise the CG. This will make the robot easier to balance. Here is a crazy idea. Set the twist of the little belts alternating between each small wheel. This will require the adjacent orange gears to turn in opposite directions to have the little wheels turn the same way. The orange gears can be made a little larger so they engage with one another, so in theory, only one need be driven from the central gear. You could have every other one driven from the center. The others can be made narrow so they miss the central gear. (You must have an even number of small wheels. Since you have 8, this would work.)

[@InYourDreams-Andia]

Cool project! Can't help thinking you'd need a physical inertia damper.. Like a damped pendulum, or whip antenna with a weight on top. Maybe you could do that in software, but possibly better to somehow engineer it in physically to the wheel? (a flywheel in the wheel?) On the Honda mono wheel, the active inertia damper is the person sitting on top. I used a passive method to damp out motion on a tall camera mast, rolling slowly on a DIY track and dolly. Worked really well!

[@Ofca_PL]

Great engineering mate!

[@reubenj77]

Would the idlers perform better if they were a bar with a concave curve to them? Maybe less slop?

[@CharlesVanNoland]

Another solution to the belt walking might be to just make the gear teeth concave, like an hourglass, so that the belt wants to just stay in the middle of the gear, rather than adding more parts to complicate things and increase the total print time. EDIT: Perhaps sticking with the belt strategy and "pinching" the gears to keep the belt centered you could double up on the smaller wheels to improve big wheel's roundness, without tapering/rounding the smaller wheels? Otherwise I was thinking somehow track the position of the big wheel to know where along the smaller wheels is contacting the ground and modulate the speed of the smaller wheels from there - which would be a bit of overengineering IMO. I'd try just increasing the total number of wheels at least by double first and if that didn't pan out then I'd resort to tracking the big wheel's rotation and compensating for the radius of the tapered smaller wheels.

[@H34...]

Flat belts tend to walk themselves out of dish pulleys, a crowned pulley makes them center but the twist complicates this, also the fact they're toothed. You can't keep a constant tooth profile on a crowned or dished pulley.

[@DemsW]

Well done James

[@YOU_screener]

you also could make a kind of coloured / white spot based circle and work with as Photo cell that tells the pcb in wich place the wheel is right now and how fast it should rotate to stay stable

[@jeffsadowski]

If you mark positions of the wheel you and have a reader for those positions you could program in the varying wheel height.

[@LakotaNativedoll]

Making the little wheels cylinders with a thick coating of super soft rubber or silicone would help since the soft rubber would compress in forward & backward motion, keeping it smooth. While also allowing for consistent diameter for side to side. ( with the right ratio of softness to stiffness). Alternatively, scaling it up while changing the little wheels to cylinders could help since side to side would be more consistant and the larger overall diameter would allow for more rollers, smoothing out forward and backward motion

[@Xlaxsauce]

they put a good bit of effort into that toy

[@mossbrucka]

I wonder if you offset the central gear down and have an idler to keep it "tensioned" if you could make the the outer wheel gears engage as they move towards the ground, that way the mechanism only drives the outer wheels when they contact the ground further reducing friction

[@adambrandt1858]

Thats so cool man. You should try the same thing but go back to un tapered wheels! You cant stop now your so close to making perfect!

[@kennethbasteiro8467]

This guy is amazing and extremely smart. Wish he lived around my area.

[@xaytana]

Tuning the velocity of the small wheels should be fairly easy. Take the position of the main drive motor, apply the pulley ratio, and define the 'steps' of the wheel, in this case 16, then alternate velocity per step, as you have eight large circumferences and eight small circumferences with the tapers. You could also refine this further, interpolate velocities between the steps by using a sinusoid that provides the function for velocity based on the position of the wheel, so if you end up standing on the middle of one of the tapered half-rollers, it'll provide velocity for that specific contact, and not just if you're standing on one full roller (two large circumferences) or standing on two rollers (two small circumferences). Also, while off-plane idlers do work, a better means of belt position retention would be belt guides. Ideally you would want the belt to be, for lack of a better term, square with the pulley it's on, to prevent walking. Assuming you can find a belt with enough flex along the axis of its width, this should be possible to do with properly modeled guides. Though this would need some kind of retention, a semi-closed channel (think of a squared C, a box with four 90° corners but an open slot in one side) could work, but so would a U-channel with additional idlers opposing the channel. This would also require low friction, PTFE tape should suffice but I'm not sure how well this would interface with the belt itself. If using a semi-closed channel, the ends should be flared on the toothed side to prevent snagging. Belt walking may not seem problematic, but there are power losses associated with it, because you're losing at least some amount of power to walk the belt along the pulley. A better fitted belt may also suffice, though. But a channel overall would be more ideal compared to just off-plane idlers. Think of a channel as a series of idlers, a constant surface easing the transition of the belt, rather than having the belt at angle A (0°), angle B at the idler (45°), and angle C (90°) on one side of the loop, whereas a channel would be a constant gradient between the two points as ∇f:0°→90° along half of the loop; yes, the actual loop is more complicated, as you have 180° turns at each end within a separate plane, but the part we're concerned about here is the planar transition between 0° and 90° along the length of the belt. In short, gradual transitions are better than segmented transitions, more efficient and with less wear assuming the same friction.

[@amirhm64]

James did it again

[@zippythinginvention]

Your parts are becoming more and more stylish. I enjoy the way you're utilizing existing patents, while deftly dancing around their violation. When you're ready, I would sure like to converse with you about the generation of unidirectional force.

[@Shokman08]

Very cool!! You are missing control authority on yaw. Ideally, you are now balancing only on a contact point. If you add your vertical fly wheel, you might have a very nice balancing system!

[@jamesbruton]

I'm investigating reaction wheels in a couple of weeks.

[@Hephera]

you could probably stop the belt from walking side to side and rubbing on the edges by giving the gear a "crown" like a crowned pully. ie making it domed in the middle so the belt wants to self-centre

[@TheMadManPlace]

think about a CV joint. fixed diameter of the outer but it can pivot to get flat on the ground, wish i could post some paint images to explain what i am trying to say...

[@SlickManChoux]

Good job. I know it's a lot of work has been done behind the scene.

[@divertechnology]

you made a design jump!! great for you

[@axelSixtySix]

Hey James, to prevent 3D printed parts to slip you can wrap around them some heat shrink sleeving. That's some quick and easy rubber coating. Regards.

[@HolyGarbage]

Since you have positional sensors for your motors, you could probably calculate the current gear ratio as a function to motor position, and thus vary the speed of the motor as it travels in order to achieve a constant wheel velocity.

[@oneMeVz]

Perhaps to deal with the different velocities programming-wise, have it keep track the rotation of the big wheel so it knows to change the velocity of the motors.

[@blinken0152]

I love this series. I'm hopeful you figure out any issues. I was thinking this would be very cool to have for a "fresh cut grass" robot/puppet.

[@maxj9204]

This looks like it would be a good fit for flywheels/reaction wheels. They could add yaw control and probably a bit better roll control too.. though getting both systems to work in tandem sounds like a bit of a task lol

[@theCombinator]

Hai James, If you would know the position of the wheel the adjustment of rotation speed for the small wheels could be fairly easy implemented. Well at least for you.

[@KaidoLP]

Maybe you can put an encoder on the forward hub and use the angle to compensate for the different diameter of the side wheel

[@RNMSC]

It may need more than a Teensy, I really don't know, but I'm wondering if you would have better luck if you had some way of tracking where the main wheel was, with a curve to apply to the amount of side to side force to apply based on that, the curve following the highs and lows of the conical sections for the small wheels. It wouldn't have to be a 1:1 match with wheel position, a 8:1 reduction would match up with the 8 small wheel locations. Part of startup would probably be walking through those steps to find max and min values to apply if you're using a simple encoder. An encoder with a endstop sensor to align with a given position on the main wheel might allow you to code in a table that doesn't have to be populated during startup, it just has to find where '0' is on the wheel.

[@bradley3549]

Super neat implementation. Smaller toothed pulleys and smaller gears would go a long way to prevent misalignment, or angle of attack on the belt. Sure, skipping might be more of an issue - but compared to the friction drive it would still be 1000% better. Smaller gears/pulleys would also allow to add MORE belts and more independently driven tires, to help maintain a more constant rolling circumference. For tensioning, that ring was pretty painful! Angled slots or ramps like you might see in an iris would have been much more elegant.

[@H34...]

If you increase the distance between the pulleys/reduce their diameter, the twist and belt walk become less prominent. This would mean a bigger wheel, but if you want to recreate honda's gadget I think that would be a plus. I think a bigger wheel would have less of a taper on the small wheels which would also help I think. To solve the side to side issues, you could add an encoder on the wheel, and use it to scale the side to side gain depending on what angle the wheel is at. You could apply a sinusoidal scaling depending on encoder position. If you gear the o drive to the wheel at the same ratio or lower as the number of small wheels, you could even use the O drive encoder output for this job.

[@beautifulsmall]

saw the tapers on the toy bike wheels, interesting, great low backlash build. Balancing , APPLAUSE, APPLAUSE

[@markstevensfpv3722]

Amazing work James!! 👊 Genuinely innovative! Please make four wheels, bust out your welding gear, and make a go kart with these wheels!!! 😆