openDog Dog Robot #12 | First Steps | James Bruton
Вставка
- Опубліковано 21 жов 2018
- It's time for openDog to take its first steps, but it's not as easy as it looks! There's going to be a significant amount of development to make it anything like as agile and dynamic as Boston Dynamics. The next steps involve new feet and an inertial measurement unit, as well as a bunch of software features and tuning.
My open source robot dog project. It includes everything from 3D printing to coding, including a lot of CNC cutting and installing all the electronics myself.
If you want your very own four-legged friend to play fetch with and go on long walks then this is the perfect project for you. This way a dog can literally just be for Christmas. The full CAD is available at the link below for anyone that’s keen to build their own.
github.com/XRobots
Get your openDog merch from: teespring.com/stores/james-br...
You can get the CAD and code from: github.com/XRobots/openDog
① GENIUS
No MacArthur grant yet. But while I’m still waiting by the phone for that call, maybe you might patronise me with your Patreon-age. Or, if you’re all out of digital cash just comment on one of my hilarious Instagram pics, pop me a Tweet or generally yell in my direction.
Patreon: / xrobots
Instagram: / xrobotsuk
Facebook: / xrobotsuk
Twitter: / xrobotsuk
Huge thanks to my patrons, without whom my standard of living would drastically decline. Like, inside out Farm Foods bag decline. Plus a very special shoutout to Aleph Objects, Inc who keep me in LulzBot 3D printers and support me via Patreon. www.lulzbot.com/
Why not join my community, who are mostly made up of actual geniuses. There’s a Facebook group and everything: / community
② XROBOTS
Former toy designer, current UA-cam maker and general robotics, electrical and mechanical engineer, I’m a fan of doing it yourself and innovation by trial and error. My channel is where I share some of my useful and not-so-useful inventions, designs and maker advice. Iron Man is my go-to cosplay, and 3D printing can solve most issues - broken bolts, missing parts, world hunger, you name it.
XRobots is the community around my content where you can get in touch, share tips and advice, and more build FAQs, schematics and designs are also available.
③ MUSIC AND SFX
The majority of the music and SFX used in my videos is from Epidemic Sound share.epidemicsound.com/xrobots - Наука та технологія
Consider using multiple files for one sketch instead of having everything in one bloated source file. The Arduino IDE makes this very easy, just click the new tab button and name it however you want, you can then move functions into the other tabs, these functions are accessible in the main sketch file the same as if they where in the main file. This will save you hours of scrolling through the big file when you can switch tabs to find the function you want easily.
To compete with Boston Dynamics it has to dance. Including moonwalking and twerking. This is very important.
we use IK in animation walk cycles for cg work, I am sure you are making it more complicated than it needs to be, How we approach it is opposite to what you are doing, we place the feet forward in space and the IK chain works out what position the limbs need to be to do the effect. How about changing the program as foot position first, then the algorithm places the limbs. That way you can figure out foot position = speed of limbs. Walk cycle also needs to be done as a pair, rather than foursome. So the front legs determine their own position, and the rear legs determine their own position, then they work together with the IMU to keep it balance.
This is the exact thing I've been thinking for the last couple of episodes.
that way you can also adjust the expected feet position depending on the IMU to adjust for tipping while it happens. right?
Copying Boston dynamics isn't so easy lol
It's called inverse kinematics. It is a lot more difficult and computational expensive than foward kinematics. Possible, tho.
@RenzVC: Yeah, that too. I don't think you can now push the dog around without it tipping over, unless I've missed some coding in this series before this. But having all four legs tied together is pretty much a very bad idea anyway. It works for manual control, but will be bad for anything advanced. A dog actually moves very dynamically with basically independent leg movements, angled legs for additional support and so on.
I really like the way James analyses a problem thoroughly before he dives in with his implementation. Much less frustrating that way.
Excellent! I glad to see that individual makers could do impressive stuff, comparable to the robotic universities and companies.
thanks!
It already looks sick with all the rotating components. Lots of things going on, it's great!
looking better and better, great work James!
TIME FOR AN SBC!!!
since you are adding sensors soon, and you need the dynamic speed control and position interpolation, it seems like the perfect time to add a layer of control higher than the master arduino.
it also seems like if you ported your geometry code (like the leg length calcs) it would allow you to redesign the areas related to the motor speed control. it just seems like this project could be a platform for so much more once it is mobile, and sensor input data can get real high bandwidth real quickly, this seems like the right time.
you can add a usb programmer so that you can download updates to it over wifi, and send diagnostics back to your desktop,
you can add a camera and stream video back to a device (laptop / FPV viewer),
the possibilities are endless! looking forward to seeing more of #opendog :)
Yep - some changes will occur after part 13
awesome! i cannot wait
0:03 grunting makes you stronger.
Awesome series btw
I knew that walking was going to be a big trick, it looks like you have a good start, you may need a tad more power to get the speed your looking for. But I feel your walking the right path
walking the right path
I don't think so. I think he has been doing it wrong the whole time, software wise. He has the limbs moving with FK, when really he should be calculating motion with inverse kinematics, where the ending foot position is decided and then it calculates the angles of the joints from there. Maybe heavier on processing, but a lot less work for him, and more reliable over all. As well, it means it can be dynamic, so later on when it needs to step up obstacles etc it can do it automatically once the spot to step onto is found.
As well, by constantly re calculating the IK, it allows the feet to stay stationary when the bot is shoved/etc. Pretty useful.
you clearly know more about this aspect of coding then I do, I will bow to your knowledge, so how would you go about coding it?
@@bruceneely4859 personally, I'd set up a sort of simulated 3d coordinate system for the bot, and send it the coordinates for its feet. If it knows the coordinates for its shoulders as well, then simply solve for each joint angle in order to get there. It's not particularly efficient, but it's entirely doable.
Again,your beyond my level there, I don't have the coding knowledge to do it, but it sounds good. I doubt that James is going to switch gears at this point, though it is an open project so someone may take on the project and get a simulated walk made unsing the cad model, while James continues his approach. I have found that he generally manages to get things working through trial and error eventually, and watching him try is part of why I started sponsoring his experiments.
Having tried to go down this path with a much smaller servo based machine, all I can say is that this is amazing.
Keep up the great work, can’t wait to see the outcome
I have had the same problem when making my own little dog robot! and surprisingly I made the same observations as you did cuz it made sense! i would suggest either increase the length of the legs or decrease the width of the main body! i have observed that increasing the ratio of length of legs over the width of the main body adds to stable walking.
Always impressive ! You are on the good way !! Keep it UP !!
I always thought that your house is just very small but this makes more sense.
James Bruton Consider using data from BLDC drivers to sense resistance/ground. It will be cleaner and easier that way. Also this can prevent robot from damaging itself due to collision.
Love the project,
(A small issue I am noticing is that those "shoulder rods" seem a bit too weak. you can see them flexing when you pause the video. Stiffer axles should help with the jerkiness also)
Keep these great videos coming :)
he already talked about it, and i agree, those don't affect stiffness
Small tipp: on every point at the walking there needs to be always 3 foots on the ground, the returning to the front needs to be twice as fast as the time spend on the ground.
Great start to figuring out how to make him walk, or at least working through the challenges you have.
AT-AT... that would make a beautifull open-dog skin! you already have the locomotion set sortaaah! ;D!
Hi James.. this project is great and you are a very inspiring engineer. When I was at the university( in the late '90s) i did with 3 other colleagues an hexapod robot. the trick to making it walk straight was to mix the signals from two dynamics : one was the accelerometer feedback to ensure that the body stayed parallel to the ground, the other was an oscillating wave to impose the gait. for us it worked very well. I will look forward to look into your code :-) good creation.. cheers from France .. divide et impera
Very inspiring!
openDog will make you swole. Being an engineer in training now, I've been more excited for this than before :)
Cant wait to see how it performs with the sensors.
For interpolation use a cubic spline function. It takes care of the dynamic aspect of getting from one point to another.
Did you choose yellow and black to get a huge nostalgia feel of the Lego Mindstorms 1.0 system? Every time I see the OpenDog, I feel like I'm watching a life sized Mindstorms bot... In a good way.
100% agreed
Great project,
I need one to carry my cooler, fishing gear, and me when I have to hike to my fishing spot. LOL.
My psudo code ish for walking: Lift front right leg extend forward. While lifting back left leg and rotating it forward rotate front left and back right. This will cause the robot to shift its center of gravity forward and have it land on the front leg. Then plant the back left leg. The repeat switching left for right.
Amazing
This is so cool, but I get a headache trying to figure it out. Great job! Thanks for sharing.
Man you need a proper workshop
It's mind blowing what he can accomplish in such a small space.
The Boston Dynamics Spot Mini's amazingness is obvious.
It can dance better than me.
Teaching robot dog to 'Do the Timewarp Again'!
Seems like getting it to walk up stairs should be a priority :)
So cool
You should get an old tire and cut the tread from it to use as the feet, of course rounded over. Tires are fairly soft and it should be able to go just about anywhere without slipping. The other thing I think the lower part fo the legs should be two peice, with heavy springs to cushion the steps a little bit, that might help stability and save the plastic part's from shock.
Others have mentioned about IK but I think you are on the right track.
Perhaps it would be nice to do a more from-the-top approach, i.e.have the legs seek a forward cg and lean the dog until a threshold is hit where the leg step is triggered in a binary way (with the stride length depending on the cg speed etc). then the step needs only concern itself with its own sequence, as you're doing.
Just some lose thoughts, great job as always, looking forward to the next installment.
As a animater put it to me once. "Walking is the art of falling and catching yourself" I feel looking at the walking program from this perspective might allow a more organic movement.
Really nice.
Suggestion - moving the gate functions out of code into arrays so your gates are data/lookups.
Your setup spreadsheet would be close to the input for such a system with the walking loop cycling an index - i.e. literally a loop rather than the state machine you currently have.
congrats
I'm wondering if you'll implement any kind of give in the legs when Opendog is taking fast steps, cause I'm imagining that when walking fast it will put a lot of stress on the robot if there isn't any give in the knees.
If I did then I'd have to measure the compliance and put it back into the kinematic model really quickly. There's some discussion on this next time.
One day that thing will walk itself up the stairs :)
Nice project. I recently completed a quadruped using $2 chinese servos. I'd say ground contact sensors and an IMU would help Opendog a lot!
that's happening next time!
One more thing is that u can add stability by shifting the Center of gravity through translating or rotating(along z axis) the heavy components(like batteries) inline with the pair of legs in contact with the ground!
Nicely done ! Can you make it run in place ?
I seem to remember seeing BigDog doing more of a 'jog' on the spot, then moving.
Nice work as always
About walking i think you have to keep the idea about diagonal foot walk.
I think you have to make the foots to move like this :
first turn the down motor in order to lift the foot and then move the upper motor so that extend the up angle , so the foot will move forward .
At the same time move the upper motors of the two steady feet so the robot lean little at front.. and then move the two down motors opossite direction so the feet will step down again.. it will be like you push forward the robot
i am not sure if i described so well so i will try to post later a drawing with what i want to say..
keep good working
This is greate. You should check out Eadweard Muybridge, his photographed animal walk cycles would be really usefull. I'd love to see it walk like a horse, a dog or a cat. I think the step order is front right, back left, front left, back right. A staggered series of steps, each returning to its start point should keep it moving continuously forward nicely. If I've worked it out correctly, the return to the start point should take the same time as 3 steps. Quick steps would alow 3 feet to remain in contact with the ground all the time.
Keep up the fantastic projects.
God... the way he says “trapezoid” 😆
Great work. use neo magnet as suspension.
Great project! I think you can ask Syraj Raval to use the 3d model to train a neural network to walk more naturally, will be a great project for both also!
To do that would require the actual mass distribution and motor torques also presumably?
In a constant state of falling and catching oneself. Kind of like the space station staying in orbit.
idea 1: have 2 feet (front or back) to have 2 wheels which are blocked (you can still walk with those). How ever, when you release the 'brake' the two wheels are running free and now you can move forward using only 2 feet.
idea 2: do a bunny walk. From base: increase the front legs, shorten the back legs. End of stroke? Now lengthen the back legs. As soon as the robot tips forward, you can start hopping forward.
Can’t wait for you to make a humanoid!
James in this video you said you are not quite as good as Boston Dynamics, I disagree they have a team of the best minds in the field working for them and pretty much unlimited funds and technology so what ever they want and need they get, you are a one man band and I reckon you would hold your own against any of the BD team.
backyardbasher he is far far far far away. Boston dynamics isn’t even that superior in my opinion.
Sorry but no. He's far away. What we're seeing here is all still analogue. What if any other situation besides our predicted environment happened? Robodog can't walk. Boston dynamics is way more advanced (like literally) compared to him.
i think his point was that given more time and funds he wouldn't do a worse job then they are... having said that the robot is really clunky in comparison...
@@im0b as far as I can see, nope. Doesn't matter with many fundings or not. He needs more higher level skills to even come close. No offense but he's not ready at all to try mimicking boston.
what kind of skills for example? ive looked into their mechanisms and it is pretty advanced but given enough funds james could hire experts, my point is that he did the 80% pretty well ...
Thumbs up m8. Awesome stuff. I learned so much through out the series.
Can we approach this in a different way?. Basically open dog is closed loop system so all the motors know where they are and they have holding torque. SO can we implement a teaching mode for motion where you move the robot to a certain position record so on and so forth. then call those motion recordings. and then you interpolate between the moves.
is it possible you could use something like a game engine or animation software to create a walk cycle with a model of the robot and then output the positions of each leg to the kinematics system? I know I don't know anywhere near enough about this. I only wonder whether it might be easier to start with something 3d, to then convert it back into code, before outputting it as 3d motion again, rather than starting with code to go into 3d motion?
My apologies if everything I have said is total and utter garbage. Loving the series though!
Making an interface to a 3d program can be a lot of work. What he has so far is a good start to doing that but he's trying to see how the machine behaves under test before progressing the software much more.
I think that's what boston dynamics is doing with Atlas on the parkour video. You can probably blend animations with adjustments for the actual environment.
That's the inevitable conclusion, I imagine. Everything seems to want to gravitate toward Autocad. First it was CNC machines and video games (3dMax is basically just modified Autocad), then 3-D printers, and now, robotic programming.
Very awesome project! I'm always looking forward to a new episode release. BTW, what kind of encoder do you use ? ( I'd like to know the brand and model please)
The ones from the Odrive shop...
Looks awesome!! Just a quick request - Can you please print new feet to the right length, it sounds like the robot is flexing and coming apart due to the 15mm shorter feet.
Move the shoulder also to effectively change the center of mass and get better stability. Another thung that i'd like to suggest/ask to you is: can't you do something like a sinusoidal function for controlling the position of the leg's tip, letting the IK model do the math for the other joints, and just offset this function X degrees for each foot that way not having to program it multiple times(?)
Please please please print some new feet! :-) presumably your model assumes a point contact with the ground, too? Fantastic video and project, thank you
yes
The easy part of the project is done. Now to the hard part :-). I think movement (agility) is the most crucial part of this project which will determine its success. I suspect a software emulation of the dog's physics is the next most important step.
It could be really easy. It's open source
Great job. Thanks for doing what you're doing.
How long does it take to print new proper length feet though? You've been mentioning these are too short for some time now.
next time, but...
was thinking tennis ball shoes with a round plastic inner for feet.
you are the best
So, the way quadrupeds walk in nature is actually very different from how you were trying to have it. They typically walk left front, left rear, right front, right rear. You can probably find some good videos of a horse’s walk gait online, and then for faster motion it will be the diagonal pairs you have, for a trot. I doubt you will want a canter or gallop gait, but those are even more efficient at higher speeds.
There was some discussion about this in the video...
Well it seemed like you were trying to do a trot gait as the walk speed?
Are the shins long enough? The movement sounds kinda painful.
this robot was born with bad arthritis
Changing gaits to predefined speeds might be useful. Just run the leg positions which would need to catch up a little faster.
Yes that can be done but they are not always fixed depending on the move.
Interesting to see it walk....
You should examine cat gaits though, since unless they're running at top speed, they are essentially placing one leg at a time, and have 3 legs on the ground. - while I don't know the implications and relative merits or problems of it, cats walk by placing one foot at a time, but they counter- place them;
eg. the sequence is front left, back right, front right, back left.
Given the secondary motion effects it's possible that's beneficial; (and perhaps works to provide stability through torso twist rather than leaning.)
... Of course, cats also place their legs in line, but this seems to be related to producing less footprints (eg. It's for stealth, not walking as such.)
I mention it mainly because ot's a prominent example of a quadruped that makes extensive use of 1/3 gaits (which can have static stability), rather than the more common 2/2 gaits that can only be dynamically stable (or even 0/4 gaits, where all 4 legs leave the ground, as with galloping horses.)
James great vid as always, I noticed that dogs paws spread out to make a wider foot when the hit a surface, just a thought?
next time!
Most dogs have tails. Robot tails can be used as a counterweight?
Looks like the first thing you need to work on stair climbing
Hello James, Great work. Development firmware complex like this must be quite hard work in arduino IDE. Check visual studio code and its arduino addon. It is free and it have included all features of arduino IDE plus you can use very powerfull code autocompletion, integration with version control and all other features of advanced code editor. VS code makes my work with complex arduino projects much easier.
Awesome as always. Are you learning a lot as you go or do you pretty much know ahead of time what your result will be but are going through the steps for our benefit?
I know roughly what needs to be done, part #13 is already made. But most of the testing is necessary so it might as well be a video.
That's about what I thought. I agree, you should make the video. I for one like seeing the process. I think the stuff that doesn't work quite right is important to put it all in context. Complicated things are created one piece at a time.
Re: competing with boston dynamics, how many mugs and tshirts do you still have to sell before you have the equivalent of DARPAs budget to give yourself?
Loving the progress! Any thoughts on how you'd make the gait generator deal with rough terrain, or is that not something you plan to explore yet?
Yes - an IMU in the body and foot pressure sensors.
I would be interested to see how you get on with pressure sensors; I have been having real trouble getting reliable readings from mine for detecting ground contact on a robot at work.
Can't you use the error reading from the motors to calculate the forces being applied to the legs, to compensate for the lack of IMUs and feet pressure sensors, to some extent?
Excellent work James! There is a pretty cool video from the royal society where they show how animal (snakes, bipedal, quadrupedal) locomotion is a wave function inverted along the axis of symmetry for the left and right side of the creature. I'll post the link when im on a pc.
ua-cam.com/video/6YVHqLE2-vc/v-deo.html
I would love to see a collab and do some ml with this hardware
Hello James, awesome creation! I am curious a couple of things-- how much power does it consume both quiescent and in full activity? And how much battery power do you have onboard? How long does it typically stay running under moderate load before needing to be recharged? (Not sure where best to reach you, here or FB-- I asked both places)
The wide vs narrow leg distance argument is a fallacy though. Only thing important in how fast sth falls over is moment of inertia. We tend to think that legs closer together causes slower falling because we mistakenly associate it with taller robots/animals (=greater moment of inertia) that fall slower.
not gonna lie James..... I well thought that dog was gonna drag you back down them stairs!!!
The code with gaitFlag1 should have been done with a switch case, and you could do inside each case a different waiting time and the transition to the next stage. Basically I am describing a state machine
I just realized it doesn't have a head haha, is that planned with optical sensors/LIDAR possibly?
how is it going to see and feel the world around it or have you not gotten that far yet?
That first attempt at walking!! I thought it was going to fall over you... Nice beast! How's your stairs winch coming along? XD
question: doesn't a dog craw/walk with the opposite "foot" and "hand" combination - so the left front and right back move synchronously?
sorry premature question
I'll be glad when the new feet are on and he doesn't have to qualify every description with, "now the math is right, but the robot is wrong..."
next time, but there may be other issues...
nylon chopping board makes carpet dragging allot easier
They grow up so fast! T.T
I've bulit a Lego walking machine fully mechanical no computers, and I have been experimenting with animation wolf/dog walk cycles. I am very interred with how they move. 😄
"James Bruton
1 month ago
it's unlikely to skip there given the reduction on the ball screw"
Hi. The Big Dog is really good a heavy duty that's why he need suspension system yup, the machine need suspension always.
But then I'd have to measure the suspension and put that into the kinematic model or the maths would be wrong.
@@jamesbruton I know it's a hard work to do and the challenge is to make the machine Walk or Run and maybe jump in the future. We all believe that you can do it.
Could you grab it by the foot & move it through the motions, record, edit, play loop ? It works for rehab & physiotherapy
bull screws cannot be backdriven
Making a smaller model which could be posed would work thougb.
The entire point of Ballscrews is to have a backlash free linear motion that can be back driven. Trapezoidal screws on the other hand can't be...
Boston dynamics make spot
James Bruton - open dog by Bruton dynamics
what motors do you use? they don't look like servos. these looks like same motors used on drones. wouldn't it be faster if it was hydraulics.
Just watch part 1/2. That's a pretty quick Hydraulic pump and sump for that much throughput.
I think if you do it fast enough, you can move 2 legs at a time by moving opposite corner together; there shouldn't be much time for the bot to tip over too much before the legs come down.
edit: I should've wait till I watched the rest of video before posting this comment...
Is there any error checking on the code , as some of the code could be alot more compact.
Using arrays or proper data structures
The amount of work in this project is amazing
(final comment) ODrive has a trapezoidal trajectory planner that may be of help in moving the legs.
I need to investigate that, but I need to move in three dimensions, so I'll probably end up doing it myself.