Optimising a 3D Printed Robotic Actuator
Вставка
- Опубліковано 21 лип 2024
- High-performance 3D printed robotic actuator, explaining key design decisions that lead to unmatched value and performance for robotic actuators under 1kg.
In short, it's a conventional 2 stage gearbox packaged in a very compact, lightweight and unconventional way.
rboticlabs.com - Specs/CAD/Electronics/Shop
Chapters
0:00 Intro
0:35 What's a QDA?
0:53 Motor selection
1:47 Gearbox design
3:24 Housing design
3:58 Performance
4:34 Extra clips - Наука та технологія
What were the advantages/disadvantages of that eccentric cycloidal planetary gearbox?
Love how you were able to identify that. I think your comment will probably be my favourite for a while.
Contrary to what's commonly claimed online, the ECC gear is not a rolling contact at all. It's only a rolling contact for a very small portion (
@@rboticcouldn't you convert it to rolling contact by using bearings on the pins? it will increase mass by a lot but should be reasonably compact still
@@rbotic haha hilarious, I was fiddling with eccentric cycloidal planetaries a few months ago, glad to see i'm not the only one anymore!
@@parkerbradshaw4122 possibly. Might be worth a try :)
the editing of this video is just pure chaos 🤣....from just removing hydraulics out of the picture to how straightforward and blunt you were with the gear assemblies....that was greta man you win a sub
Amazing video and a awesome detailed explanation. Very usefull and entertaining too! I'm looking forward to see more of this!!
incredible work, can't wait for your channel to take off with more content!
This is amazing! Well done!
This is really amazing stuff, can't wait to see more!
Would love to hear more explanations of concepts for those of us who are not experienced in this field, looks very interesting!
this is unbelievable. amazing job
Not a second wasted! Rare gem of a channel, had to subscribe. Funny when you pointed out the low waste on the 3D print, so disconnected from the objective
Great project and a very well explained video! Would love to see more! And would like to see details about the material used and the printer settings.
This is amazing. Great work. Can't wait to see more!
More to come!
Thank you for saving me months of experiments! Gears it is.
30 seconds in and i've subscribed. You're doing a great job. keep it up
Awesome video. Loved hearing your thoughts on the various drive mechanisms. You gained a new subscriber! I'm curious to hear what the torque efficiency, backlash and wear/tear looks like. I'm building something similar (with the exact same motor) but with belts.
Thanks! ~81% of theoretical torque at all tested torques. I expect the transmission efficiency to be 90%+. Backlash is negligible imo for legged robots, you could always preload the gears if you require zero. Wear is visually zero. Belts are a great idea! It'd be more shock tolerant, quieter, but I'd expect it'd be a bit bigger, heavier and more difficult to assemble. Depends what you value.
Subscribed!
I've been building my own BLDC motors, axial flux ones. One of their advantages is more efficient use of electromagnetic flux -a lot of the flux in these radial designs seems to go straight out the side of the motor, Hallbach array on the magnetic circumference would probably help concentrate magnetic 'juice' inwards where it can do more work...
Looking forward to seeing more of your robot and what it can do!
Glad you enjoyed. Update video soon!
Can't wait to see that robot walks around :)
pretty awesome, can't believe youtube just now recommended this video when all i've been watching is stuff like this....
yo the amount of stuff you had to have expertise in is kinda mindboggling to me, sensors/navigation + material sience + mechanical engineering + software development + control systems + genetic algorithms(?), insane (for a dude with a 3D printer)
Absolutely wonderful. I just got my first resin printer. I can't wait to start making this stuff! Are your designs available for download?
I applaude for lightness, efficiency, simplicity, and force!
Just a thought, put up your design to some world wide shipping website/manufacturer so, that instead of CAD-files, we could buy the product. =)
At some point I'd need a robot for experimenting and I like your design goals. Big up from FI!
You can buy the product but currently is not in stock as far as i saw
The actuator will be available within a week from now ;).
Great video! This is now listed as my fav robot leg design. At the end of the video, you are training the robot in simulation. Mind already telling us what program you use for the simulation, i.e. PyBullet etc? And the algorithm you prefer.
Thanks! Issac gym. Reinforcement learning PPO is what it uses.
@@rbotic Sounds great... I am building something similar to what you build here. But, instead, I am trying a cycloidal drive and also using wood for the frame. Isaac gym needs NVIDIA RTX GPU as I know, and I don't have it. So, the simulation I do will be on PyBullet
Great work. I want to see more of it in action! Are you still testing and working on programming a run without supports?
Stay tuned 😉
awesome. nice design! i'm working on a 1kg barell gear that can output 200Nm with petg cos it's forgiving of printing imprecision. You can use steel tubes inside plastic gears and angle grind a criss cross pattern in them and then use a glue to glue the plastic gears onto the steel tube, the gears need a stellar or internal gear pattern to fuse properly, i dunno what the best glue is.
fascinating
Nice build! I am curious to know how the backlash of motors translate to the foot or ankle. You can lock the rotors and push the ankle to measure. It is okay to have some play tho if you are doing force control. Also, when you do locomotion control, the robot will definitely freak out at one point, and if it hits something hard (like ground), it could break weakest point along the transmission (usually the pinion). Under 1kg is absolutely amazing, but I am curious to know if it would survive real world testing.
Since it's preloaded, there's little to no backlash. Surprisingly there hasn't been an actuator failure in the months of testing. Though it recently snapped its own leg off at the hip during a sim-to-real transfer freak out 😆. Happy holidays!
This is soooo cool 😮😮😮😮
Sick stuff! Do you find that having that much torque makes it easier to design a motor controller? Of course, it's back-drivable, so I imagine that would also make it much easier to do IK and whatnot.
Actually, speaking of which... that software sneak preview with the different agents all running around. You wouldn't happen to be working on a learning-based agent, would you?!
I'm super stoked to see where you take this project!
Thank you! A bigger motor generally means a more expensive controller, doesn't influence difficulty. It is a learning based agent :)
This guy out of nowhere... great actuator design, very fair price too
Subscribed ❤
I don't have numbers of the competing actuators to compare with yours but I have a feeling they are not as much worse in strength to weight as you might think. Usually industrial actuators are not rated at peak torque/power values the way you are rating yours - they usually have a significant margin of safety. I'm sure your actuator costs a lot less to produce in low volume, and its impressive what it can do. But I have my doubts about longterm dimensional stability of plastic gears, and the space efficiency lost compared to other actuators.
Thanks for watching! Definitely agree with you on the space efficiency. QDA are almost always thermally limited, and their torque is constrained by that (rather than mechanical).
This is really cool. Would love to see more. What sim is shown at the end of the video?
Thanks! Dw there'll be more. The Sim is Isaac Gym.
We use conventional cycloid drive gearboxes on our robots (FIRST Robotics Competition). These are almost fully 3D printed in PETG but we do use steel pins on the perimeter and for the output. Our current robot has a couple of 24:1 Cycloids about 4" dia and a 48:1 that's about 6" dia. We use two rows of Delrin balls (1/8" dia) at the perimeter to provide a smooth rotation (instead of expensive X-bearings). The cams run in small, deep groove ball bearings. The secret to their success is that we use three cycloid gears in each gearbox, each spaced at 120 deg to the other. This provides almost 100% tooth contact and eliminates vibration. Have you tried anything like this? I suspect that our 24:1 could handle more torque and looks to be about 1/3 of the size of your actuator. The downside is that it does have many more parts. I love the simplicity of your design. Be glad to share if you are interested. Cheers.
I'd be more than happy to share designs. Please contact me at rboticlabs@gmail.com.
Yes I've tried a triple disk cycloid (blue and grey design in the video) . I wasn't happy with the weight/complexity/total size with motor. I never did any proper torque testing as it was 3 years ago, so your cycloid could possibly output more torque.
I don't like the expensive 4 point contact bearings as well. I've been looking to eliminate them but can't think of a good solution unfortunately.
I wonder if you'd be willing to share this via a video? or do you have open source cad files for this? extremely impressive design!
@@gedr7664 Are you by chance FRC team 7664? If so, I can point you to Chief Delphi.
@@derekfinch9586 I'm not, i'm from the UK and we don't have that here. I just build robots for fun. Would it still be possible to get an insight into your design? :)
Loving this project; what kind of drivers, controllers, and protocols are you using?
A COTS DGM controller using can bus
nice !!!!
That is a really cool actuator, good work! Just one question what backlash does the gearbox have? From what I know cycloidal, harmonic and belted gearboxes are specifically used because of their low backlash characteristic. Multi staged geared gearboxes, especially 3d printed ones, usually have quite a lot of play, which disqualifies them for a use in precision robotics.
Its negligible. You could preload the gears to get zero backlash at the expense of some friction. But imo it's unnecessary for legged robots.
@@rbotic Yeah I guess you're right, low backlash is much more important for automated robotic arms. By the way, what type of sensor are you using for position estimation? Is it a rotary encoder, hall sensor or something different?
Rotary absolute magnetic encoder
The training video at the end 😆
Hahaha. Hey it's an attempt alright 😁
I know bldcs have pretty high specs on paper but aren't like BD hydraulic actuators actually a lot stronger in terms of peak usable power?
I mean like they make a giant metal robot jump and do cartwheels...
Not to throw shade. You are doing awesome work!
Interesting.
cool actuator, mind sharing the print files? im sure i could model something myself but i'm just interested in experimenting for now so it would be a huge timesaver.
Thanks! CAD is available on my website as mentioned. I don't think I'll be sharing the print files for that reason. I've shared how it works in the video, so you can always design one yourself! :)
mmmm i m gonna follow this, i m surprised by the torque / speed ratio
Ok you are amazing. Keep up with the good work. I would loooove to see a video of the AI simulated world u teased a the end of the video. Please show this walking
I wish you showed the complete assembly with motor, housing etc. Will have to look At your website
You are selling the 3D so we will never see how to assemble le the inside.
It is what it is. At least you now know how it works 😉
Nice!
How do you mount the motor? inside the ring?
How do you then get the output to rotate?
Was thinking of making something similar but belt driven instead.
Also, how powerful are your motors (as in how may watts, amp and volt they run on)?
It's mounted to the housing with 3 screws inside the output gear. Belts would be better for sound and backlash, but I wouldn't recommend it for other drawbacks. The specs are on my website.
if you buy the files is there a video or a tutorial on how to make it the actuators?
That cycloidal gearbox with the rollers looks very interesting, I have seen similar systems in the past but no matter what I google I cant find another example or explanation. Do they have an official name? why did you not use it? what was its flaws?
I don't know the official name. But it requires very high manufacturing tolerances as well as stiffness. Much more than your traditional cycloidal gearbox. When loaded the rollers would slip as the plastic flexed.
Thanks for the feedback @@rbotic knowing why all your failed gearboxes well failed is almost as valuable as seeing your successes
Awesome! Bought one.
Why not vesc controller?
Thanks you so much! I believe you're the first 🥇. Vesc controller wasn't designed for robotics, although it would be possible to use one. Main reasons are, VESC doesn't have an integrated absolute encoder, an optimised CAN bus, and it's quite big.
@@rbotic thank you !
💖💖💖💖
How did you replace the shaft of the brushless motor to power it from the other side?
How do the reduction gears stand with long time usage? Have you noticed a lot of erosion?
I've noticed zero wear with months of use. Since they're plastic their typical failure mode is overload or heat.
After making a bipedal robot with the actuators you designed, will you make a similar quadrapedal robot? How about a similar humanoid robot?
No plans on making a quadruped robot. But there are plans to make arms for the bipedal robot. Thanks for the interest!
That's fantastic!
what driver are you using? Odrive? I would love to see the training possess are you using Isaac sim because I could never get that to work properly?
In terms of carrying capacity, a multistage gearbox is not as good as a planetary gearbox. What do you think
Torque capacity is comparable to a planetary gearset of the same diameter (assuming
Very much loving your designs! I've yet to venture into dc motors because of the expense and complexity with the actuators, but using servos has so many disadvantages in comparison, you've helped inspire me to take the plunge soon! Any plans to release the cad files or are they available somewhere?
I'm glad you're inspired 😊. There aren't any plans to release CAD files at the moment.
Sensored or sensorless brushless motors? I have issues with low rpm cogging even with high reductions. Brushed motors have max torque at low rpm but there are other drawbacks
Sensored brushless (abs magnetic encoder). Allows for max torque at zero rpm. Downside being more expensive than brushed unfortunately.
@@rbotic true, ESCs are pricey, thanks for the explanation
Holy shit!
😂😂😂😂😂😂This guy makes some crazy claims lol!
Are you using some sort of custom FOC controller for this or some off-the-shelf part?
Its a COTS DGM controller running custom firmware
@@rbotic amazing man the system looks great
That actuator seems crazy strong ! , Definitely what is needed for a competent legged robot
do you know how much backlash these actuators have? Thank you
So what is the max torque? And how much weight can that arm actually lift? I feel clickbaited, the thumbnail and all the leadup and no actual numbers...
All the specs are on my website as (briefly) mentioned at the end. Hope you still found it entertaining and informative on how it works/was optimsed. Core specs are 47Nm peak @ 997g.
Oooh! Instasub!
I’m designing a robotic dog. Might use something like this
What outrunner motor is that?
8318. Actuator uses 8308
This is probably how IG-11 works
Here's a cool fact: humanoid robots should do 100ds of different tasks, tea, washing, carrying, tidying... Garden robots just need to do 5 jobs that machines already do, digging, sowing, weeding, cutting etc. sooooo... what is a more useful investment of time? the one that feeds 8 billion people with 50 cents of fruit and veg a day or the one that already has competition? :) hi from the UK.
Idea is great. Can't it be made smaller?
Thank you. I've made a smaller version, but wasn't able to hit the same torque density due to the reasons mentioned.
What is the ratio of torque between both versions? Also what about using other technologies like resin printing, cnc machining, sls plastic and metal? Will they increase the torque of these actuators?
@@shivstark8494 similar ratio. Funnily enough using another manufacturing technique wouldn't increase torque. Reason being is that the limiting factor is thermal not mechanical. You may be able to make it a bit lighter though
4:34 "B-BAKA!!"
oh dear isaac gym
I do think it's currently the best option for accelerated RL. Any suggestions for alternatives?
no it definitely is the best for sim-to-real RL. just based off my experience both getting a stable walking gait and good sim to real transfer is sorta a nightmare though.@@rbotic
Look, as long as its not going to the Titanic, you should be fine.
Well, the BD gave up on hydraulic atlas so you planning on competition with it by proxy now won the contest against electric atlas.
im honestly disheartened by their decision due to love of hydraulics and how intricate they are
Are hiring interns... I'm interested
List your projects, so he can see
Amazing video and design. Thanks 🎉
PS: I’d recommend a more catchy name and website. It was practically impossible to find using google. Also related video (and a catchy name example):
ua-cam.com/video/cqpYpj8EaDY/v-deo.html
Thanks for letting me know!
I agree with you , it was hard to Google the website , a name change would be helpful
Use the power of AI to make it walk efficiently, also add some feet so he can walk as a test.
Lightweight✅
Durability ❌
It’s still durable for plastic
The key frame says "designing plastic to rival steel" - what plastic did you design? Did you invent a new fiber-reinforced resin? Which one is it? A lot of us ME's would love a plastic that rivals steel! Is it glass-filled PEEK or maybe carbon-filled Ultem™?? Come on, we're all dying to know what resin competes favorably with steel! Spit it out!