openDog V2 Part 1 - Robotics R&D Design Process

[@jamesbruton]

Patrons and Channel members already have more previews of this project. Also please check out my previous research using back-drivable brushless motors to make compliant joints: kzbin.info/aero/PLpwJoq86vov9Aogc6PseW8BUe2nQyZESZ

[@Nae_Ayy]

OMG you released it early love you

[@abrarmohamed3351]

Can you please make a video of how to code a robot dog 🙏🙏🙏

[@Mitiya__ua]

Can u also highlight cost of materials for each project and amount of time it took. Curious what effords for u it takes to make one build from scratch to finish

[@nathanchalecki4842]

Does this outrigger style of motor handle side loads as well as a normal style motor? So excited

[@MuhammadDaudkhanTV100]

Excellent

[@Skyentific]

Wow!!! Thank you a lot for mentioning my channel!!! I am really impressed!!!

[@sofuckingannoying]

Funny how a line of text can have a deep Russian accent.

[@20vK]

Your channel is absolutely fantastic. It deserves a mention!

[@Marci124]

It's nice to see you two drawing conclusions from eachothers projects.

[@MiniLuv-1984]

Subscribed to your channel. Funny thing about youtube, you can subscribe to many channels...it makes little or no difference to James, possibly enhances his standing (that is already at epic proportions).

[@elminz]

You got an under appreciated channel, been subbed a while now.

[@JakeHillion]

Your R&D process has been so satisfying to watch. It's seemed subtle over time, but every iteration and change has so clearly taught you something. Love these videos and really pleased to have finally joined your channel.

[@jamesbruton]

Thanks for your support Jake!

[@1894db]

Well you haven't fallen for the Sunk Cost Fallacy, I"ll give you that. Looking forward to V2!

[@HugoPeeters]

I love that you show more of the design process, the rough CAD and the thinking behind it. Thanks James!

[@tomtronic]

Really nice way to start a new project.

[@jamesbruton]

thanks!

[@20vK]

I agree - such a great way for people to follow the project. It also refers to previous videos, so should be good for James's channel overall. I think the design process is fascinating for lots of people and will only add to the relevance of the project for a wider audience. It allows for such a greater understanding of the challenges and the solutions or compromises James decided on in the development of a final product.

[@alexk4709]

I love that I didn't pay attention to this project for a bit, and when I come back there have been 5 different iterations built that are all unique and working. Pretty dang impressive 👌

[@theRedCat5014]

Love seeing the entire design process, can't wait to see future parts!

[@jamesbruton]

Thanks, I'll try to keep the detail there in the series.

[@habakkuk414]

Yes, I have really missed this in your videos. I enjoy the R and D thoughts as much, if not more then the engineering.

[@RENbocK]

The amount of work that goes into making even a minute of footage is mindblowing. Seeing Boston Dynamics come up with this was sci fi, but they have an endless pool of resources. now its already in the realm of hobbyist's?! Your an inspiration

[@alessi4249]

I do forget how much it costs to do a project like this until I'm reminded of the fact there's 12 motors £80 each haha I'm at least glad I can witness the project. Great vid

[@Marci124]

Same here, I'm planning a project with three BLDCs and I never really internalised the cost.

[@ciarfah]

Never mind the cost of all the ODrives on top of that

[@Mike-oz4cv]

It makes you understand why he’s advertising all the merchandise and his patreon account.

[@pixelfairy]

thats what simulation is for

[@michaelprice3031]

Glad to see that you are headed in the right direction. Keeping the weight toward the center and using high torque motors is the way to go (or at least the way I'd do it). I think you should make the legs less bulky though. you don't need all that material for it to be strong enough and it will move faster. Besides that, I think you're on the right track.

[@syninys100]

In terms of bulky legs: The important parameter in terms of affecting speed is the moment of inertia (the rotational analogue of mass, hence ideally minimised); and the actual volume of the leg is not relevent to that. Mass further from the axis of the rotation has more impact, so this is one reason why many legs taper. This needs to be balanced against the stiffness of the leg - and that's governed by the material choice and the second moment of area of the leg (ideally maximised). So maximum stiffness for minimum weight is given by a large and hollow leg. This looks bulky, but if it's as empty as possible in the centre, then it's actually a very mass efficent shape. Until the part starts moving fast enough for air resistance to become an important factor - but we're not talking about a propeller here! Anyway, the design looks to me like the lower leg would be very easy to iterate through a few options on, once built, and that's the place where it would matter most, so I suspect that it's been designed to allow rapid iteration, and hence deferring optimisation on that part for now.

[@dekutree64]

From what I can see on Boston Dynamics and MIT dogs, they generally keep the foot under the hip axis, but do so by rotating the hip. Seems like a better solution than building an offset into the lower leg. If nothing else, it just looks more organic/less rigid. When using IK on the foot position rather than driving joint angles directly, it doesn't really matter if the shoulder/elbow axes are parallel to the floor or not.

[@baldomo11]

They also use torque-controllable motors to enhance the "fluidity" of the movement i think

[@ukrainewarroom]

They look so "alive" it's creepy, the Boston robots are in uncanny valley.

[@rbrookstx]

I’m always amazed James doesn’t have way more subscribers. I’ve been around here for years and I’m always inspired to push myself harder when I watch his stuff. Thanks for doing what you do man. It’s a favorite of mine and my daughters.

[@Whiteoverred5]

Love the new style of showing design process

[@doughntworry]

Your videos are amazing. Very clear and super informative. They have taught me a lot, not only about the technology you use but also the way you prototype and iterate. Showing the progress from the ground up was super interesting in this video! Looking forward for more!

[@DiabolicMagicSquare]

Great.. Can't help but get inspired by people like you. It is one thing to innovate but selflessly put the knowledge on open source is commendable. Knowledge is far more valuable than money.

[@RobertKvsv]

By the way, thanks for guiding us through your design thought process!

[@thomaswasselin6335]

It always impress me how much work you do between two videos.

[@CantonGregory]

Funny to see Skyentific appear in your video. He just started appearing on my feed last week.

[@DantalionNl]

I have seen these motors in a 7 dof robot arm and they are amazing so fast and yet so precise. I think they will be an excellent choice!!

[@colsoyeti1]

Great to see the design phase of ODV2, I feel so inspired after watching that I too want to take up cad lessons to further my 3D designing. Keep up the impressive work James and loving the new T-shirts.

[@nathaniellangston5130]

I love the new dog design and seeing the process! I'm glad you mentioned Skyentific! I have been watching his channel for a long time and he has fantastic stuff and nowhere near enough subscribers for his content quality!

[@dooby1445]

Funny, I was just trying to find that wide diameter motor for a design I want to play with. James, what you're doing here is seriously one of the best things in the robotics community.

[@nanorex_designs]

Very happy to see one using a backdrivable brushless motor setup. This one looks promising!

[@Abdulrahman_Nosser]

extremely happy with the new style of videos please keep them coming

[@KnightsWithoutATable]

Great summary of the project and the direction it is going after the R&D results.

[@photon2724]

wtf. Someone fund this man! He's a genius!

[@photon2724]

I really appreciate that you don't give up! You make it better every time!

[@Jonnymiller1977]

Ive been enjoying your vids for a few years, blown away by your engineering skills, but this is the video ive been waiting for. I really loved seeing your design process, starting from a known component and working your way outwards. Great work

[@DrMake-qx3gk]

Nice to see more of your design process and thoughts about the decisions! Can't wait to see a version of openDog getting stable to build my own at some point.

[@multiplio2924]

Thank you very much for going through the design process! It's really interesting how you get to that super detailed looking CAD at the end, and I'm looking forwards to the next stages!

[@sammflynn6751]

These videos are gold. A small suggestion, If your controller isn't reacting fast enough, maybe look into a controller which uses a combination of feed forward and feedback control. Relaying only on feedback control is slow because ,for the controller to correct the plant,error has to occur first.

[@DerIstDerBeste]

The word 'feedforward' is new to me. Do you mean something like a (e.g. laser) distance easuring device to find out the distance of each leg to the ground so that it doesn't get rammed into the floor? I think that might help, but the robot should still be able to cope with big errors through feedback. That's especially necessary for unpredictable ground (like items on the ground, gravel or elastic/soft floor). There's always gonna be errors, would be great if the robot could cope with them.

[@MrPmjg]

@@DerIstDerBeste no. Instead of waiting for the error if the PID to grow, you can add another term that is based on theoretical calculations ie: how much current do I need to provide do the motors to reach a target speed? You can start to feed that value in the beginning instead of waiting for the PID

[@sammflynn6751]

@@MrPmjg Nicely explained sir

[@ukrainewarroom]

@@DerIstDerBeste proximity sensors? 🤔 they are about £5 each and used on CNC machines and 3D printers etc.

[@MrAquaktus]

This is amazing James! The speed of iteration is astounding, the likes of which I have seen very few achieve.

[@queabbs]

Love it. I am into the detailed walk through on your design process. Thank you!

[@RonaldFinger]

Ayyy Skyentific is great! Also, looking forward to seeing the development on this project!

[@jreneew2]

Robotic Fiero??

[@pugglez4798]

I love this open dog series. I also like you talking about the design. Keep it up 👍

[@antonwinter630]

thanks for going into the extra detail of the design process. hopefully by moving the motors into the shoulders and driving everything by belts will give it lots of torque!

[@Flux9665]

Up to the point you've shown in this video, the legs could just flip upside down and still work exacly the same. That could be a feature, if it falls upside down, the front legs rotate around the front, the back legs rotate around the back and the top is the new bottom. On a more serious note, really cool ideas and nice to see the design process!

[@isthissiddh]

i really liked how you explained your thought process for the design. please keep doing that.

[@FusionSource]

Hey James, love your projects, thanks for sharing again, always excited to see what you going to do next.

[@ovidiurosu6632]

Just amazing. First of all, great work with explaining your line of thought. I find it easier to understand and I get more practical knowledge out of it (as oppose to the wow factor that can't get me started). Secondly, your focus on research and explaining what you learned from each version is humbling. I'd like to follow your example and do the same with my projects. Thirdly, I can't comprehend how one person can do all this in such a short time. I've seen teams with 10+ members that have half your output. What's the secret? Thank you and keep up the good work.

[@slimknight_]

When discerning the pulleys hitting the legs, you could move both pulleys to the outside so that they stick out a little bit in the front and the back, leaving you with an even load. Great work!

[@jamesbruton]

I'll probably just cut the back off since they don't need to rotate round and round.

[@s.k.t.3855]

Hi, James Burton! I was looking for robotics KZbin channels and I just couldn't find quite the right content, that is, until I found your channel! I'm a student who is getting their second associates degree in robotics technology, and I absolutely fell in love with your channel when I saw your channel trailer. Subscribed immediately, keep up the amazing content, and absolutely cannot wait to see openDog V2's next video!

[@TheJohnestOfJohns]

Following these R&D series has ben incredibly entertaining! Great work and best of luck.

[@ukulelefatman]

Thanks for the insight to your design process. Really looking forward to the next iteration.

[@jobadro]

My brother is pretty much building this exact robot at the moment. He also uses pretty much the same motors and also the belt system of the cheeta mit robots. He just ordered some more motors and odrives.

[@brenno3735]

TWO WEEKS?! I can't wait that long!

[@MatterLabz]

I really enjoyed seeing your design process. Very interesting.

[@_Garm_]

this format is much more interesting to see and follow :D

[@vondarycrentsil9180]

Dis is the best robot dog I have ever seen , u are the best, keep it up

[@nathanskinner423]

It really seems like the motor could face the other direction. The encoder could go on the outside or be driven by its own small pulley but them being inverted is spreading the mass out a lot.

[@akselwilliamdanenbarger7969]

Thanks for teaching how you think. 👍🏻

[@WistrelChianti]

Great to see more time spent on the CAD and thinking process again ^_^Y

[@simondibbern5564]

I've followed your robot dogs attempts from the beginning and always wondered if there might be a fundamental issue with processing power in your designs. Standard Arduinos are nice and accessible and all, but the performance can be pretty terrible. Higher end hardware is not that expensive, compared to the rest of the design. I would recommend to do some investigations or experiments on latency - all the way from the sensors to the motor drivers.

[@jamesbruton]

I'm using Teensy 3.6 which is a 32-but Arm Cortex M0 running at 180Mhz. On top of that each ODrive has another STM32 all running in parallel dealing with motor velocity/position control. SHould be quick enough for basic stability control, and it seems to have worked out so far?

[@simondibbern5564]

@@jamesbruton Fair enough, I must have underestimated the Teensy.

[@zerobow9413]

@@jamesbruton Check out the Teensy4 if you need more juice

[@Joelmelanie]

wauw James i loved wathing you yourney. such passion into this project🥰 all the best Joel

[@steffen1182]

I am always deeply impressed by what you come up with. I am happy about every video and can easily follow your explanations and thoughts. Thank you for all the work you put into the videos and for the countless hours of work it takes to produce these videos. Unfortunately I do not have the skills or time to implement such projects.

[@PeterKlein0064]

Why don't you angle the (blue) block holding the motors up a bit (have the hip pivot point up by ~20°)? This would give more freedom of movement to the front legs but keep the motors inside. this way you use some of that movement potential (360°-Pulley) above the mechanism. It should not shift the center of gravity too much.

[@RexusKing]

The new motors look so sexy! I'm just happy to ride along the development process. Sorry couldn't make any donations though🤗

[@ajaycycle78]

Bhai aapka bahut bahut shukriya yahi to dhundh Raha tha main😊😊😊

[@idkgaming9978]

You should make a quadcopter that somehow drops a rover of sorts that can then scout out an area. This could be useful for things like mapping out mine fields if you give it the ability to detect mines and mark where they are.

[@aw_dev]

You make amazing videos! Did you notice that there are 2.1 thousand likes and only 19 dislikes? 👏

[@hairdinischmietz1801]

Awesome build

[@mostafafawzy2081]

Loved the idea of details ❤️... thank you very much for your efforts

[@ammarkov]

Looking forward to this build.. This one is going to work!

[@chrism4621]

A possible fix for the big pulley on the hip. Use an intermediary pulley. A small pulley on the motor, going to a larger pulley. The larger pulley has a smaller pulley attached, this pulley spins freely. From the small intermediate pulley a belt to the same size pulley on the hip. There is some extra hardware, but you move the big pulley out of the way. This project has inspired me to build something similar.

[@jamesbruton]

I'll probably just cut the back off because it doesn't need to rotate round and round.

[@boblelan]

Keep up the good work, you really inspire me!

[@nagualdesign]

Now you're talking. Boston Dynamics' motors are more disc-shaped than cylindrical, so these new motors certainly look the part.

[@10mrmarkin]

Turnigy should sponsor this guy

[@AllenKll]

Is it really necessary to use the same motors for the hip joint? The hip joint should have relatively small movement, considering it will cause a greater effect per unit of movement. Maybe something smaller would work? I know it might complicate the control model using different motors, but all joints don't have the same requirements, so why use the same motors?

[@IamNomadSauce]

Well, I just found my new hobby/obsession

[@RobertKvsv]

That merch looks really nice!

[@nophead]

As your joints don't need to do a full 360 you don't necessarily need closed loop belts. I built a solar tracking bot where azimuth was a closed loop belt but altitude was a length of belt with its ends fastened to the large pulley which wasn't a complete circle, just enough to give 90 degrees of motion. Another possibility is joining the belt with a tensioner at a point where it won't hit either pulley like a 3D printer.

[@jamesbruton]

yep - the hip joints will probably end up with the pulleys cut off, as long as I can get the motors in the right place so that I can move the joint enough.

[@avigetsbored]

Might be valuable try doing a topology optimization for parts of the legs, and the frame, it would help to lower the overall mass and if you lowered the electronics as low as you can on the frame it could help you get a mechanical advantage as far as balance. The lower you can move the mass(without moving the mass onto moving parts of the legs) the more balance you will design into the robot and the less you will have to compensate for an unbalanced structure on the computation side of things.

[@poyot8643]

it will be more cool when you can control the robot far away from your location ,you can buy groceries without leaving your home,that cool 👍🏼

[@DavePetrillo]

It would be cumbersome, but if you printed the most distal 'shin' part as one piece and wanted to put your belt in, you could use tree supports for the overhangs of your print and pause the prints and set the belts right in there and finish the prints with them already around the pulleys. Would make a very slightly stronger and lighter part.

[@ryanruthrhona]

Excellent video. I think you should start a separate channel that shows/teaches the masses how to start using Fusion 360 and the ins and out of 3d printing but by having a project oriented approach. Due to all your background in robotics and mechanics I think it would be a fun way to learn.

[@thomasre8073]

Thank you for you detailed insight.

[@greatmatze]

my question may be dumb, but do the pulleys have to be round ? when they are just rotating a few degrees they would clear the legs if they were flat on the outside?

[@jamesbruton]

Yes I'll probably cut the backs off them...

[@magnussorensen2565]

19:46 soo freaky. He just pears over the orange edge.

[@10p6]

I like the idea of having all the motors in the shoulders, however. Weight could be reduced and the whole thing simplified if you stack the motors next to each other, and then use the motors case to drive the belt directly instead of using a pulley. It would be possible to also stack the hip motor next to the other two, so all the weight of the motors is directly over the top of the leg making a very slim profile.

[@Marci124]

I was wondering if just directly interference fitting a sprocket shell on the case with some driving dogs would have some drawback, it doesn't seem proper somehow. But why would it, the case is directly coupled to the axle and has the same runout and adequate stiffness. The surface also falls between the bearings, should be good enough.

[@andrewsnow7386]

He's hoping for a 5:1 reduction in the drive. So, if he drives the belt from the large diameter (92mm) of the motor case, then the driven pulleys would need to be 460mm (18 inch) in diameter.

[@10p6]

@@andrewsnow7386 You equation would make it much more than a 5:1 ratio. Also, he wanted to put some spring in its step, my system does that naturally.

[@andrewsnow7386]

@@10p6 The specs I found say the motor case is 92mm in diameter, and based on the video that looks about right. To have a 5:1 single stage reduction, the diameter of the other pulley needs to be 5 times as large. And, 92 X 5 = 460mm, further 460 / 25.4 = 18.11 inches. So where am I wrong?

[@andresvanegaspatino8642]

Excellent video! Can you please tell me which modeling software you show at minute 12:29 to which one it corresponds? Thank you. Excellent work.

[@UFO_researcher]

The latency due to processing sensor data is a big issue. You need to calculate the force of the motor from the torque using the equation 'T = F * r', where 'T' is torque in Newton meters, 'F' is the force in Newtons, and 'r' is the distance from the center of the motor shaft in meters. For example, with a 4 cm diameter primary gear, 'r' is 2 cm, T = F * 0.02. If you are using a spur gear, the rotational force is equal to F*cos(angle), where 'angle' is the angle of the spur gear (usually 20°). 'F' is force in Newtons, however, the velocity and distance of rotation is determined by the RPM speed of the motor.

[@mossm717]

Looks like there are loads of interesting servo quadroped projects on thingiverse, but most lack sensors and stability control. The most advanced one Ive seen is the opencat project. Really impressive walking ability. They have a tiny one that uses 9g servos too: ittybitty kitty

[@charetjc]

18:45 any particular reason you don't want to reverse the forward knee joints to clear the shoulder pully?

[@MuhammadDaudkhanTV100]

Fantastic video

[@AJB2K3]

Bicycle puncture repair kits are useful for creating closed loop belts.

[@richardlincoln886]

I know you're prob sick of back seat drivers... Ideas - thoughts - pulleys vs. bell-cranks for knee joints due to limited travel, might also let you put both shoulder and knee motors at the shoulder?

[@Sharklops]

Not too big, not too small.. you should call this one Goldidog

[@stealthBae]

Nice project you got! I would suggest to increase the frequency of your main loop. This increases the stiffness and responsiveness. If you would like to do that with Arduino's I would use (multiple) ESP32 microcontrollers. They run on a frequency of 240 Mhz compared to regular arduino uno 16 Mhz.

[@jamesbruton]

I'm using Teensy 3.6 which is more than fast enough, on top of hat each ODrive has an STM32 dealing with the motor velocity and positioning.

[@stealthBae]

@@jamesbruton Nice!

[@Zaze09]

Love the channel and want to support. My kids would love some robot dog t-shirts. How about adding some kids sizes to the store?

[@kremsnita446]

A robot dog army!!!

[@davidbarrett7424]

Skyentific warned against small pulleys due to slip problems. Go with the planetary reduction in the shoulders James learn from MIT / spot mini etc. Think the body design will be more compact, none of those clashing issues either. what ever you choose wishing you good luck with the project!

[@devdsp0]

It looks like most of the robots you're referencing have a hollow axle for the shoulder joint and run a solid axle through it for the knee joint. Using concentric axles would allow you to have both the motors on the inside of the robot's frame. Is this something you considered?

[@GreenJimll]

Could the hip motors be mounted inwards using another belt drive to a pulley on the hip itself? That might help with the clearances?

[@Pentross]

possibly save weight if you have super high torque “waist” rotation rather than 4 independent leg “hip” rotation?

[@reggiep75]

I don't know what the weight bearing tolerances of the dogs lower leg but with it being a bit spindly (on initial looks) and at an inturned angle I'd be inclined to infill part of the print with more plastic or put a thin aluminium 'bone' in them and give it a wide ankle & foot. Model the foot on a cat or dog foot and possibly use springs to ensure the floor striking part of the foot is at the right angle.

[@maxtorque2277]

Make sure you account for the motors rotational inertia, especially when referenced by a gear ratio! Do the maths, and you'll find that you'll use most of the motors power to simply overcome its own inertia! The higher (numerically) the reduction ratio between leg and motor, the more this comes into effect. What you really want is a direct drive motor, that only moves at the same velocity as the leg, but of course this then means your motor must have sufficient base torque to deal with the limb loads directly. There is a trade off here. Before you design anything i would suggest trying to set some base values to use as target setting. Get the rotational inertia of your motor rotor (either from the manufacturer, model it in CAD, or use a "trifilar table" to actually measure it) select a sensible maximum limb acceleration, velocity and load, and calculate the limit masses and ratio's. You will find this drives your mecahnical design! If you look at the existing "high performance" robots, note they all have very "spindly" legs, ie low mass, and use quite direct drive actuators (low interia). That is not by accident.........

[@alexdavies2036]

Have you considered using stallgaurd-capable trinamic stepper drivers? I think they should, theoretically, be back-drivable.

[@CRISCOL64]

Cooooool!, Keep going! I want to see all development 😀

[@TheAnachronist]

James, do you know of something cheaper than the O-drive but with similar functionality for much smaller BLDC motors? I'm thinking of doing some dextrous robotics with highly backdrivable motors.