Best BLDC Controller: ODrive vs MIT Mini Cheetah vs Moteus (MJBots)
ฝัง
- เผยแพร่เมื่อ 24 พ.ย. 2024
- Three brushless controllers with the same motor: comparison of the low speed performance of the ODrive, Moteus (MJBots) and MIT mini cheetah controller. I hope this will help you to choose best controller for your robot.
Special thanks to my special Patrons: Wesley Corp, RobotDigg Shanghai, Pierre Freyssinet, Shounak Bhattacharya and Tom Rodinger!
Please subscribe. This will help me to develop other projects like this, to bring the bright future closer!
One time donation:
www.paypal.me/...
If you want to help this channel, please support me on Patreon:
/ skyentific
Instagram: / skyentificinsta
Twitter: / skyentifictweet
Facebook: / skyentificface
#DIY #Brushless #Robotics
We need a list of recommendation books for the theory of BLDC motor controller
Thanks for the great comparison!
However, I don't believe the moteus low velocity performance was compared using similar gain setups. ODrive by default has a non-zero velocity integrative gain, which decreases torque bandwidth but increases smoothness of low speed maneuvers. moteus by default has this gain at 0 in order to maximize torque bandwidth. You can enable the same gain in moteus by setting `servo.pid_position.ki` and `servo.pid_position.ilimit`. If you tune those, I was able to achieve basically identical low speed performance between ODrive and moteus.
The audible noise is a different matter. The majority of the noise in a default moteus setup comes from position noise reported by the magnetic encoder. I set up an ODrive with the same AS5047P as on moteus with an mj5208 motor on both and got similar audible noise volume, although the spectral content was different likely due to the higher control rate of moteus. Exactly what magnetic encoder did you use in these tests?
I came here just to see if Josh would respond and as I expected he did. I think Josh is the kind of person that is both very knowledgeable and someone that wants to make the best possible product. I'd like to see the Moteus tested again with the adjustments and encoder as suggested.
@@wgm-en2gx I'm here for the same reason. I'd also suggest checking current shunt resistors on all the boards to be the same, as on Moteus they are tuned for high current loads, such as 60 ampers. Obviously, during low-speed operation, you will have a bunch of noise in a current loop.
With ODrive I used: as5047p used with a, b signal.
@@Skyentific Thanks! My test was with SPI, guess I'll try A/B/I soon here and see what results it gives.
What does Oskar Weigl of Odrive say to this? Yes, I also believe the noise as well as the motion (un-) stability comes from different PID settings.
Thank you for demonstrating the characteristics, I’m experimenting with MJBots and had the same results, I thought it was me doing something wrong.
and? how is MJBots?
This is so cool! High accuracy BLDC is so useful
Smooth motion during calibration is related to the fact that BLDC is controlled in open loop for calibration purpose (it should be, because no precise speed is needed). Problem of BLDC motor control at low speed is caused by low quality construction of BLDC. These motors do not have exact sinusoidal current waveform (compared to AC servo). BLDC motor has a noise which is added on 1st (fundamental) harmonic (as I can remember noise appears at 5,7, and 11 harmonics). AC servo motors do not have problems with low speed control because AC servo is designed and constructed in such way that current output response is pure sinusoidal (no higher harmonics that cause the noise).
Hey love your content, you gave me some real good insight into BLDC motors with your videos. Just wondering if you could make a video demonstrating how to use the anti-cogging feature on the Odrive.
You've got a new sub ... I like the DIY robotics and controls for BLDC motors :)
Thank you and I hope for more of this.
Thank you!!!
The smooth motion during calibration is due to it being done openloop with relatively high current. It is simple to do this but wastes energy. The difference in noise is likely due to pwm switching frequency. You may want to move above audible range but then your switching losses increase.
They all operate with the switching frequency above 30kHz. There is some complicated mechanism behind this noise, which I do not understand.
Thank you for the comparison. For me, I tried Odrive with D5065 motor with low speed and it seems that, it have the same problem of MJBOTS controller with low speed. My opinion I think it's related to the motor also.
Very useful comparison, thank you! I was thinking about switching from ODrive to MJBots because I thought MJBots would have smoother motion and not as bad cogging, but now I see ODrive is already good. Like another commenter, I originally assumed the noise was due to PWM frequency, but you indicated that they were all above 30kHz. I find it also strange considering ODrive control loop frequency is 8kHz whereas MJBots appears to run control at the full 40kHz PWM frequency, so we would expect tighter control with MJBots and less noise. Interesting.
Thanks again!
The odrives control loop is 8khz but the phase PWM is 24khz
You can't really expect smooth control at low speed when you send position targets.
The boards probably have a velocity mode that can be smooth even at low speed.
The calibration phase on all these boards is probably exactly the same method, it runs in open loop (foc algorithm) it doesn't use the encoder readings. One reason it looks smoother on Odrive is probably only because they use a higher current.
Without this smooth control you cannot use them for precise motion for the direct drive or for the camera gimbal.
Don't forget the noise! The ODrive is super silent. So, I think there is more than just current value.
@@Skyentific I agree you can't use this position loop for a gimbal it's better to use torque commands and have external control loop. About the noise that is curious, maybe the pwm frequency is not the same on all boards. You said there is no encoder on the Odrive board? This means the motor is driven with a different algorithm it could explain.
All these board have open source firmware and documentation the difference is probably explained somewhere. You can ask on their discord also :)
@@boblelan If the Odrive is using a better encoder, like the CUI AMT 102, it could explain the much quieter sound. You can also reduce the sound by reducing the controller gains, at the expense of performance
I think a fair comparison would involve evaluating the anti-cogging features on the Odrive and MJ-bots controller, plus running through the recommended tuning procedure on each motor with a significant (camera sized) load attached.
@@calvindekoter2128 exactly you're right! And there is maybe an anticogging mechanism in Odrive that isn't present in the others.
In this video I used the Odrive with the magnetic encoder (as5047), similar to the Moteus encoder (as5147).
ODrive have posibility to compensate coggin torquq. You need to calibrate motor for cogging torque compensation to be able to youse this functionality. The noise is because of PWM frequencey. ODrive uses 24kHz PWM by default. Seems like other controlers use 12kHz or even less. Thats why you can hear motor noise. The second reason I think is that ODrive firmware uses 50% PWM at all three phases as the ZERO vector reference, and other use 0% PWM duty cycle as the ZERO. Thats make difference on low torque. I can illustrate it with pictures if you would like to know more. P.S. You can write your own firmware for all of controllers:) Hardware is simple STM32 with motor driver, difference is in firmware.
I would love to see links to the motors
Love what you do
I am quite sure it's the Turnigy Multistar 9235 100KV but I would love to see the link too :)
Can you please build a BLDC Driver from scratch/opensource on your own. it will be a boost to bldc community.
check out simplefoc, i think thats what you are searching for
@@jan.feucht Yeah, there are already lots of people developing open source controllers, but none of it is very accessible to us noobs. A tutorial on how to design a board to run SimpleFOC would be extremely useful. Maybe 4 parts, 1. how to select components and design the schematic in KiCad or other free editor, 2) how to lay out the board in the editor, 3) how to order the board from NextPCB, 4) how to configure the firmware. Part 1 is the most important. There are plenty of existing board designs to look at, but nothing explaining the purpose of each component, so it's difficult to learn how to design your own from them.
@@dekutree64 Thats an easy question, checkout the datasheet of the main component, normally the manufacturer of the IC describs in his datasheet where to put which additional component. If you dont know how to design a board, then start study electronic engineering, if you only want to use them then buy it. For example the Simplefoc board uses the L6234 from STM if you search for the datasheet you will find one with 14 pages, on page 7 is a Schematic Circuit with the components. Datasheets are sometimes more useful than a book. The other questions sounds more like lazyness 😉
@@jan.feucht I would've asked the same question but it's not laziness, just the chaos of being self taught. I found your answer to be really helpful too
Awesome content! You are my go-to guy for Odrive knowledge. Could you do a video on tuning a motor for Odrive and running the motor without first running full_calibration_sequence? Thank you. :)
i just ordered 2 flipsky 4.20 vescs to control my hoverboard motors on a custom art project im working on
All the motors you use are cored bldcs, at low speed/power the induced magnetic field on the core (cogging) has a significant torque compared to to the control current. Gimbal motors are coreless, so they do not have cogging, which makes FOC at low speeds much more even.
non homogeneous movement in slow motion is caused by motor cogging torque (caused by inconsistent magnetic flux in motor) - to eliminate this, regulation loops have to be very fast and properly tuned or utilize some sort of feed forward. Also position sensor must be more precise - 12bit (4096 values/rotation) results in very small speed resolution. Best way to remove cogging is stator construction redesign (special motors for precise movement) or operation at higher speeds - rotor inertia smooth this torque ripple
Thank you for showing this. I have been investigating the same kind of problem with brushed DC motors and have been looking into how encoders perform. I think the problem here is depending on the encoders performance and how they are used. I can see, that at least some of the controllers, that you show, do use an encoder, but do they all three got an encoder? One step could be to record the encoder type on each case - I should like to know that.
When you read the signal from a magnetic angle encoder, you need to be aware, that it measures the magnetic signal at its own measuring cycle with its own cycle time at about 100 us. This can create variations due to how you interpret time between two measured angles. When you read the angle with only a little time difference, then you will get a substantial relative noise in calculated difference angle of the two very close angles.
I recorded a similar video of very low speed on a DC motor with another micromotor as encoder: th-cam.com/video/Sp55jztsI80/w-d-xo.html
This was really Interesting. I have found many servo systems work better under load, otherwise they get jittery where the motor moves quick and the encoder is constantly trying to correct . I wonder if you did the same test with simulated load
You forgot simplefoc 😄
Excellent information, thank you 👍😊
Great Video! So usefull! Thank you for sharing such information.
Just a small question, what is the adapter that you are using from USB to CAN? Because there seems to be very expensive ones.
I am more interested in the one used for the MIT mini cheetah controller.
Hi, the problem lies within the feedback from the motor.
Using an encoder with very fine increments is the best way.
If you have no encoder, the only feedback you can get is from the field of the motor magnets moving next to the stator windings. This is a very weak signal when the speed is low, which is the reason why the controller will have a hard time to get an accurate feedback.
You have the same problem with RC cars when using brushless motors without encoder on a rock crawler where the motor is only turning very slowly.
Awesome video where are these controller and Bldc motor use??
The noise difference is most likely because of the different switching speeds. I wonder if all of the controllers are FOC or the jerky ones are trapezoidal.
Are there tuning parameters that can improve the low speed "homogeneity" of the movement? I have several motion control projects that I need good universal BLDC drivers for, one should function more like a servo so I am leaning towards the O-Drive but would like to explore lower cost options as well!
Thanks
I wonder if you could tune the PID values harder than just 2x to get better behavior on the Moteus?
(Also, no-load behavior is notoriously harder to control than motor with load.)
ODrive рулит. Дорого только.. Хорошее видео,спасибо. Блин не могу отделать от ощущения что он щас скажет что нить типа "Балин,подшипник кривой в моторе" )))
1st. I would go ''trinamic''.. 2nd. U need a gear reducer..maby a cyclic gear.or planetary... 3rd. it's best not 2 have 2 many motion control OS'es..but a motion controller. that works best! good luck with IT !! in the future will we see U with an oscilloscope??
I wonder, could you review the Open dynamic robot iniciative? I have been locking at it but idk what type of driver they use, also I am very curious by the fact that they are usong drone rotors as motors.
Bonjour, Quel est le modèle utilisé chez Odrive le S1 ou le pro, merci. Hello, which Odrive model do you use, the S1 or the pro, thank you.
Super...Thanks.
А как насчет старых контроллеров от Алекса Моса? Не знаю реально ли их еще купить. Но все таки они для подвесов разрабатывались, и прекрасно работают без подобных рывков
Curious if you’ve seen this one: th-cam.com/video/OZvjfbpXpro/w-d-xo.html
Her motor controller is called the NearZero and specifically designed with home robots in mind.
excellent link thanks.... this place too!
How did you get to run the motor at low speed with the mini cheetah controller? Cause I tried adjusting the velocity but the motor always goes max speed. Is it true that the mini cheetah controller only offers torque control mode?
Awesome job dude.
exactly what i needed! what about the vesc? it has servo functionality too!
Unfortunately, I never used VESC. But I've heard that it is very similar to ODrive.
@@Skyentific Id love to see your opinion on the vesc and the difference between the OEM Trampa Vesc 6 and the open sourced variant similar to the Flipsky or Unity Vesc's. some are quite pricey, but some are quite cheap, so it may be a good budget option.
@@Skyentific Нет не очень они похожи. Если vesc перевести в режим управления позиционирования то он работает таким образом. Например если двигатель находится в позиции 30 градусов и нужно перевести в положение 300 градусов он просто передвинется по короткому пути. Тоесть пойдет путь не 270 градусов а в 90. Это решается кастомными прошивками которые имеют кучу своих особенностей. А как контроллер для привода к скейтбордам он совершенен. Очень странная реализация FOC алгоритма
did you try odrive with anticogging?
Great question. I used ODrive 3.5, I think it does not have anticogging
@@Skyentific It does! 3.5 and 3.6 both support all of the firmware features. The newest (best) anticogging on ODrive is currently only available on the "devel" beta branch, however. If you hop on the Discord we can show you how to enable it if you'd like.
@@wetmelon7409 Which discord?
I hadn't even realized that the ODrive HAD anti-cogging until I read your comment two weeks ago, totally changed my direction on my robot design. I am grateful for where your comment took me. I was so blown away by it in fact that I made a video of my own to explain it to others. th-cam.com/video/CWzuBoYLISY/w-d-xo.html
@@AndyVickersNet your video was very useful for me and a great explanation
100k here we go!
noise = frequency so a low frequency will have a lower hum, higher frequency will have high pitch and then if you go over that it will sound like nothing (bats would probably not like it haha but a mic wouldnt pick it up and we wouldnt hear it)
Could you check out slotless bldc motors?
The noise is generated by PWM of the controller. Probably, ODrive has higher PWM frequency, which makes sound go to ultrasonic range, but heats FETs more. It could be easily fixed with firmware modification by increasing timer frequency.
What about features? Which controller would you prefer if someone would order a robotic arm from you?
I don’t think that the noise is from pwm. All these controllers use frequency higher than 30kHz. And, if I am not mistaken, ODrive uses lower frequency, but it is silent.
Ideally would be to have moteus controller with ODrive performace. :). It is easier to work with moteus, but odrive gives better performance. So I would probably choose ODrive (but for some projects, ODrive is too bulky).
@@Skyentific So, the ideal solution is ODrive, but for single motor and designed on a round PCB with magnetic encoder, like MIT Cheetah?
Do you know which mechanisms are performed during calibration?
Have you compared with TMC4671?
What is the development kit that you used for this test?
Is it possible that the encoder is generating this effect?
They all use similar magnetic encoders
From my experience, the encoder is very likely to be the issue here. You have noted that the motor is running smooth while calibrated and the reason ist because of is running open loop. Once you close the loop, any nonlinearity of the encoder will directly transfer to unsteady motion. These low cost magnetic encoders are very sensitive to mounting inaccuracy. Try adjusting the actual distance of the magnet and wonders might happen ;) If you like to test a high end servo drive and encoder made for robot actuator applications, please let me know.
The problem in low speed closed loop control is the low resolution of the magnetic encoders ;)
grate JOB Skyentific!!
Haha I got tasked dealing with this problem once on a system that had 2cpr it was a real pain
do you have a video where you assemble the odrive testing device? how did you manage to mount the encoder to that motor?
how are you reading the distance covered by the motor? are you reading 3phase? do you have an arduino interface or this?
I think this is software related and also due to the windings of the motors; they are not equal in terms of resistance and magnetic field being generated. The calibration algorithm is flawed and the ramping is probably too crude.
there is a difference btween given torque and given speed maybe try how they perform under slow speed with addig a resistance like grap the stator by hand
very interesting content !
There is a software update that solves this low-rpm cogging problem, I've made a video about it on my channel
Nope. I've made the firmware update! :)
@@Skyentific The 8318 behaved like it was before the update, the test stand motor behaved like it was after the update.
I saw your video, before making my video. ;) That why I updated the firmware to the latest, before the test. Do you have noise from motors?
@@Skyentific noise is a separate problem, noise did not chnge with te update, but the low vel performance got way better.
Круто, как всегда!
Спасибо! :)
That's not a fair comparison. At least gain values are not equal and current shunt resistors in each phase are definitely different.
Where do you buy these motors and what are they called?
Can you share me the link where you buy BLDC motor of your clip?
Do these motors run in open-loop mode?
can i get a, schematic from this controller opensource ?
i love like this videos and your videos best.
Show the graphs!
Where can you get the mit mini cheetah controller ?
Does the ODrive use an optical encoder instead of magnetic? Perhaps it can do tighter control without losing stability if it has a higher precision encoder?
All uses magnetic encoder, including ODrive. All encoders have similar characteristics.
@@Skyentific Ok, good, then that's not a factor. Thanks!
What about Tinymovr?
maybe Unscented Kalman Filtering is needed
what about vesc how does it perform?
Hello @Skyentific. I have been following your videos on motor controllers and brushless motor. Thank you for comparing these 3 controller. Could anyone direct me how to build or where to buy the mini cheetah alike or other good controllers. I have been looking to make 10 and I have not been able to find a cheap way to do it. Any comments or suggestions is really apprciated.
How much kg can this lift from your experience?
have you tried the dev branche of odrive firmware with anti cogging?
No, but I should! :)
@@Skyentific yeah, definitely. I tried it with a rather problematic (high cogging moment) BLDC and it did improve the smoothness quite a bit, I'd say it eliminated it completely. it's only a 64mm diameter motor, so it's harder to see, but the plot looked smooth.
Вери пйорфект) Гуд стаф. Спасибо за материал!
Весьма познавательно. Спасибо!
What about the VESC?
Odrive controller replace the VESC
I think the MjBots controller is almost the same as the MIT mini Cheetah controller. Only the MCU is different, as is the firmware
Yes. But firmware is extremely important.
@@Skyentific Absolutely! I'm going to discover the difference pretty soon I guess(I am working on a robotics project at college). Luckily the code for both are open source (as you showed in your earlier videos), which can act as good references.
The MIT Mini Cheetah controller might have problems with the PWM frequency and not so good capacitors resonating, which is possibly making the sound while running.
Check also VESC :)
Yes! I'd like to see this too.
Why would you use this motors in slow mode. afaik the cheetah controller was explicit developed to react fast with fast movements. guess the others too. so comparing apples with bananas might be interesting but useless
can anyone suggest a motor control for hoverboard wheels. I have four wheels I want to control.
Odrive has a section for hoverboard wheels. But it really depends upon your application.
You didn't show us all data so we cannot judge.
You use same encoder on each test? Always magnetic encoder? Maybe magnetic interference from motor to encoder. Try optical encoder?
All use similar (but not the same) magnetic encoder. as5047, as5147, ma700. Why to try optical encoder, when ODrive works fine with magnetic?
I never planed to make this video. :)
@@Skyentific To learn! To rule out magnetic interference to e the encoder. :)
I wish that turnigy wasn't sold out
Been trying to get them for weeks :(
I don't want the best, I want the cheapest thing I can find on Aliexpress that has precision control with FOC.
moteus is an 170$ controller, what do you mean it doesn't work?
motels is 84$ controller. And it works, but it has its own limitations.
@@Skyentific I see. how about Ben Kats's controller, how much were 10 pcs you ordered from nextpcb?
@@Skyentific here is a present th-cam.com/video/NsvH4ZVp4tE/w-d-xo.html
a comment! ;)
👍
I usually love these videos, but this entire video I was cringing. This should have only been posted for viewing after all the manufacturers had been contacted about the results. Only then can you say what they are capable of. Otherwise, it's just a different form of an unboxing video. In the real world, there are soo many tradeoffs when tuning things, these motors cant come stock tuned for each and every single application. Most high-performance items require tuning to achieve their peak specialized performance.
What is homo-genius?
Жаль я уже заказал Moteus контроллер. Пожалуй останусь на Odrive. У миничиты контроллера у меня просто горит постоянно драйвер и документации по нему почти что никакой. Odrive во всём идеален.
Мотеус в принципе не плох. Если вас это утешит: я заказал 7 мотеусов для руки, а потом увидел этот глюк на малой скорости. Но думаю где-нибудь их применю. :))) К тому же, чел который их разрабатывает, очень быстро и качественно все делает. Так что я не удивлюсь, если через пару недель он все эти проблемы решит в новом firmware.
Спасибо за ответ. Да конечно исправит, я тоже так думаю. К тому же я в Китае заказал уже на производстве печатные платы. Очень вдохновляют ваши видео потому как сам делаю робота по чертежам inmoov, но использовать сервоприводы мне как то не по душе.
@@pagetheme Вы живете в РФ, сюда доставляют moteus ?
@@germankulit3548 Нет я живу в Украине. И нам вообще ничего не поставляют :). Но по исходникам из GitHub, в Китае на заводе jlcpcb, заказываю платы и детали на AliExpress. И паяю в ручную. Так и живём :)
Зря Вы переживаете. С моей точки зрения мотеус все же сильно выигрывает по сравнению с конкурентами. Просто его не совсем честно здесь сравнивали. Контроллер оптимизирован под другие режимы работы. Взять хотябы сопротивление токового шунта в каждой фазе которое очень мало для токов которые имеют место быть в видео.
Ну, хотя бы субтитры на русском вставлял бы )))
wow these motors are not designed for that at all! im surprised to see them move so slow and smooth. must be wasting HUGE amounts of electricity. very stupid
Yes! All people in robotics are so stupid! Why they all use brushless motor?!
@@Skyentific oh dear ... let me explain... these motors are supposed to run at few thousand rpm. that is optimal electrical efficiency. for slow precise movements it is best to use some gear reduction which also increases torque like harmonic drive but you would need very large gear reduction for these motors so gear becomes heavier than motor. even small gear will be big improvement. i havent watched your videos yet. give me time
@@NSAwatchesME I think the theory is that, if you can do it all electrically, then you can save cost, weight, space, and complexity that would have been used for a strain wave or cycloidal gearbox, and you'd also save backlash if you'd have gone with the cheaper planetary reduction option. Especially weight, but also space, is at a super premium on any kind of walking/moving robot.