3D Printed Robot Dog Climbs Over Obstacles
Vložit
- čas přidán 27. 03. 2022
- AD: Get your 45 day trial of the Arduino IDE for Visual Studio: www.visualmicro.com/page/Ardu...
It’s the next part in the series for openDog Version 3. This whole project is open source, and the CAD and code are already published on Github. All of the mechanical parts are designed to be printed in PLA, and the main motor drives, of which there are twelve, are cycloidal drives which I developed over a few videos last year, each drive has two cycloidal discs and a lot of bearings.
There are three videos in my channel about building the project already which you can check out, but in this video, I’m going to make some minor mechanical changes, and attempt to make it walk more dynamically.
The main mechanical change I’m making is to the lower legs. I talked about the inverse kinematics in part 3 of the series.
This involves being able to position the foot in XYZ cartesian coordinates, and work back the maths to calculate each of the three resulting joint angles per leg to achieve the required foot position. To move the foot on a path we interpolate between the positions from point A to point B which means we basically scroll through all the positions on the way, so the foot moves in a straight line.
At the moment, the dog is walking by moving the feet to fixed positions on hard coded timers. The time it has to take a step gets longer as the step distance increases, but it has no sense of balance.
CAD and Code: github.com/XRobots/openDogV3
You can support me on Patreon or buy my Merchandise:
***************************
Patreon: / xrobots
Merchandise: teespring.com/stores/james-br...
***************************
Affiliate links - I will get some money of you use them to sign up or buy something:
***************************
Matterhackers 3D printing supplies: www.matterhackers.com?aff=7500
Music for your CZcams videos: share.epidemicsound.com/xrobots
***************************
Other socials:
***************************
Instagram: / xrobotsuk
Facebook: / xrobotsuk
Twitter: / xrobotsuk
***************************
CAD and Code for my projects: github.com/XRobots
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 Lulzbot, Inc who keep me in LulzBot 3D printers and support me via Patreon.
HARDWARE/SOFTWARE
Below you can also find a lot of the typical tools, equipment and supplies used in my projects:
Filament from: www.3dfuel.com/
Lulzbot 3D Printers: bit.ly/2Sj6nil
Bearings from: simplybearings.co.uk/
Lincoln Electric Welder: bit.ly/2Rqhqos
CNC Router: bit.ly/2QdsNjt
Ryobi Tools: bit.ly/2RhArcD
Axminster Micro Lathe: bit.ly/2Sj6eeN
3D Printer Filament: bit.ly/2PdcdUu
Soldering Iron: bit.ly/2DrNWDR
Vectric CNC Software: bit.ly/2zxpZqv
Why not join my community, who are mostly made up of actual geniuses. There’s a Facebook group and everything: / 287089964833488
XROBOTS
Former toy designer, current CZcams 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. - Věda a technologie
Man this guy is a super talented engineer.
With money. 😅 It's pretty experience to build this thing
@FelinoFacho He kicked catdog !!!
*skilled. Calling people talented downplays the time and effort it took to achieve this level of skill in any area.
@@needamuffin true
Skilled*
A bit disappointed to hear that there won't be an open dog v4, but still eager for your projects every week!
sounds like a community project to me
although I love OpenDog and yes it's sad that there won't be v4
he still has options to revisit for 3.5
me personally am Subbed for the E-Bird videos hahaha
cuz that's what I want to build after he finishes the final version!
OpenDog has come so far and it’s definitely been one of my favorite projects to watch on CZcams
When real dogs walk, they use a 4 step sequence. Have you ever considered using a modified 4 step sequence vs. the 2x2 ? As fast as your joints move, it seems like you could reduce the tipping, or swaying, by only having 1 leg off the ground at a time. The modifications you made in this episode are amazing! Such a huge difference!
James's project is fully open source but the only one who can fully assemble it is him. Well done!!!
Now I want to build it, when you mentioned this.
Then again with the learning which comes from building open dog, One could gain good expertise in building robots.
@@Abishek_Muthian Couldn't agree more 😄
If I understand correctly, part of the project's goal is to make it possible for anyone to make this robot. Hence, saying that the only one who can assemble it is him is calling that part of the project a failure.
To that end, I think he's done a very good job on that goal. It's all off-the-shelf parts or FDM printed parts - even the cycloidal drive. All the off-the-shelf parts seem relatively easy to source and relatively inexpensive.
I could build it. But I don't have the money 😓😅
Yes, this is what I needed at 1:00 AM, it’s not like I have a paper to write
There’s only one paper that can’t wait, and you’re not putting anything you want to read on that one ;) 💩
OpenDog is such a good project. Awesome work James, keep killing it!
awesome woof
i thought you were gonna say keep Kicking it 😂😂
you couldeven say its a good doggo.
And im glad james is giveing v3 plenty of walkies
13:33 "I don't really want to..." Glad to hear you haven't put an end to the evolution of Robot Dog. Many thanks for the inspiration. 😃👍
There just looks to be so much opportunity for more evolution here, I'd love to see version 4, or higher!
What he's done is nothing short of amazing from a technical and gee-wiz point of view, but at the same time, the capabilities of the legs seem no more than could be achieved with a wheeled robot (MUCH easier!) the main reason to use legs like this is to be able to handle big obstacles, things like walking up a staircase. This robot isn't there yet, but I can see how it could easily get there with a bit more tweaking!
From what I've seen he seems more interested in making OpenDog a way of demystifying the sorts of robots Boston Dynamics have made, and making a technologically close equivalent, but built at a budget that doesn't make military contractors cry. Sure he could burn another small mountain of time, cash, and effort making a V4 that'd maybe have some better specs, but with the project being more about making it have stable-self regulated motion that adjusts to it's environment and has enough power to move around without cheating with bungees or using rigid code with no IMU home boy has knocked it out of the park, and now with it released into the wild I expect we'll be past the "LOOK I DID A SPOT" clones stage in a year or so and have folks doing some Real Cool Shit with OpenDog derived stuff in short order. Good on ya James!
My girlfriend walked by and asked "Why is that robot walking on magic wand tips?"
I said "for friction" without thinking and we both laughed. 😂
James making open source sex toys when?
@@GingePlaysMinecraft Isint that already a thing? Just take the leg part, adjust code and add off balanced motor inside the magic wand tip...
Everything is a sex toy if you try hard enough.
@@Hellsong89 I'm pretty sure the print lines on that flexible TPU wouldn't feel great on your junk...
@@Hellsong89 I assume it is. Wouldn't that fall under rule 34? Lol
Re-watching this 8 months later I'm still sad that there won't be a version 4. I loved watching the project develop and I think there is further development that could be done, but I also get that it could get stale
I would really like to see a project where instead of classical tuning and control, Machine Learning would be employed.
With most of your projects being described in a full 3D model, existing ML Models could be adjusted and trained on virtual simulations of a robot and then used in real life!
Not very easy to tell the AI what is "walking properly", even if, I guess, it could be done, by combinating horizontal speed and keeping the back flat as a "success" indicator.
@@GirardNicolas Yes that's one of a possible option of the basics of training AI. I think though that it's ripe now for James to try his hands on it, as not only has the science come far but open tools are available and this exact problem has been tried and solved multiple times in different ways for years now. Of course, it's up to James if that is interesting to him.
This is very related and shows how it is possible: czcams.com/video/knIzDj1Ocoo/video.html
And this is the cutting edge: czcams.com/video/XM-rKTOyD_k/video.html
And their robots have similar issues Open Dog V3 has like offset feet from shoulder joint.
Do you think a Jetson nano or other embedded ML chip would work, or would he need a dedicated computer to run the quadruped algorithms? I love this logical next step in opendog v3
@@richardsuckerson49 The hardware needs to be powerful enough to run the already trained algorithm. This usually scales with amount of data. With only internal data like the around 50 analog signals in the paper from 2020 it shouldn't need overly powerful hardware and Jetson Nano or other specialised hardware would probably even be overkill. Microcontrollers like most Arduinos don't fit for another reason though, the tools to run an algorithm need to be present for the platform.
Anyway, it definitely seems like classical tuning is hitting more and more diminishing returns in projects, that's why it came to my mind. Also, state machines came to their limits for robot kinematics years ago, so I would advise skipping trying to go in that direction.
This field can be a bit daunting if the tools and platforms are new to someone but most of them are pretty modern and there's documentation and communities and James already dived into that direction a while ago with vision applications.
@@GirardNicolas There are existing papers on this, it's been done quite well. Check out Two Minute Papers' channel if you want short summaries.
TODO: Growling when being kicked.
This reminds me so much of the Boston Dynamics dog, definitely up there in my top 10 favourite projects you've done. I hope he's a good boy and gets lots of hugs.
I feel like this sets a really good foundation for a AI robot dog, with the addition of some cheap sensors and cameras, you could do object recognition, avoidance, you could potentially even make it jump over obstacles if tuned properly (however dimensions may need changed a bit for that). Honestly, I would definitely get into this project myself if I had the time or knowledge.
Except the BD one moves like a real dog, or more like a graceful cat, instead of a drunken sumo on all 4's.
Its as if a whole department of professionals worked on it using careful maths and sensors to help it other than just stabilization.
Bro come on, if you can't see the similarities you need to re watch video, this is a community project, it gives people a good start to improve on it.
@@EnsignLovell Sure, fine, but no need to fangirl nutsack shine over the shortcomings and put it on a par with a $70k professional product.
If you review the video "Robot Dog Autonomous Functions you can see that's the way he is heading. That project adds in a Jetson Nano which provides some AI functionality. It could be augmented by addition of further sensors such as Lidar, GPS
I've been following this and your other projects for a few months now. I absolutely love all of your work! Its really impressive. Keep up the great work
This is so, so incredible. The most impressive 3d printed project I've ever seen.
this is absolutely phenomenal your work is incredibly impressive and the fact it's all open source is truly amazing
My OpenDog keeps hovering around the back door like it wants to leave.
This is incredible! The performance on natural surfaces looks really close to commercial robots! Congrats!
Remarkable and inspirational. The work you have done on your projects, has pushed the boundaries of what most people thought, could be done with 3-d plastic printing and off the shelf electronic modules. Your thoroughness in creating proof of concept and prototype models is exemplary.
I'm amazed at how you manage to keep improving your robots, fantastic designs!
I don't know if you did it on purpose but, it was cool the way it managed to step to the beat of the music. I've never really heard somebody explain with such great detail, every movement takes a lot thought. Thank you.
Brilliant!!! I’d love to see an Open Dog 4 though! :)
Been following the project since the beginning, I can't imagine how satisfactory this must be looking back. Amazing work!!!!!
Do you mean satisfying? Satisfactory means good enough but not particularly notable
Amazing work as always James 👏 Would love to see further development on open dog, it’s really cool!
People are amazed by James's engineering skills and consider him a super talented engineer. They praise his open-source project, OpenDog, and express how much they enjoy watching its progress on CZcams. Some suggest improvements, such as using a modified 4-step sequence for walking and incorporating machine learning for enhanced capabilities. They appreciate James's work in pushing the boundaries of 3D printing and off-the-shelf electronics, finding his projects remarkable and inspirational. Many express their excitement and admiration for his continuous improvements and impressive designs. Some mention the resemblance to Boston Dynamics' dog and express their desire to see further development. Overall, they find James's work incredible and commend his dedication and skills. That said comes the BUT! We are All waiting for a OpenDog Version IV
This really highlights just what an engineering marvel the human body is. Even leaving out organ function, just the kinematics alone are pretty awe inspiring
A lot of work has been done for this project. Respect
Your 3D printed builds are truly mind blowing.
It was nice seeing the project, can't wait for the next one.
Looking forward to next project. Great job so cool brother.
your videos make me look forward to every Tuesday!
This is truly incredible. Amazing work.
I think you could really nail the last few key details in one more iteration! I know these dogs are extremely challenging, but I vote you do one more hahaha. Wonderful work!!
starting to look alive! impressive work. =)
Insanely cool vid as always good music too
An excellent finale. Let’s see you in the next season project!
Impressive and a wonderful color too!
You know one day it's going to kick you back.
Great video. Keep up the good work.
You are awesome James!
This walks better than any of your other dogs! Looks great!
That's an impressive project. Well done!
This is amazing work!
Awesome Work!
Excellent work!
I love watching your videos I learn something new every time
I just want to cheer for the little creature every time I see it move!
This is a fantastic project. It's starting to enter the uncanny valley where we start to judge it as a living creature which ought to have better control of its limbs. Now it seems clumsy, nervous, unaware of its situation and inefficient in its movements.
I think you have good enough sensors and mechanics, but I think the software needs attention next.
Why not start again almost from scratch with the software. remember the lessons you have learnt, factor out some fundamental things into common code, but start again with a new motion idea.
Two sources of inspiration come to mind:
1) chickens and other birds have an amazing ability to keep their heads in one place. Explore a motion idea that aims to keep the position of its sensors as still as possible. e.g. when you lift one corner, it tries to make itself heavier by not pushing on that leg, and when that is not enough it straighens the opposing legs.
2) Cats seems to be the most sophisticated and efficient movers. Start again again and try to behave like a cat. This time optimise for battery life and energy expenditure. That bent leg pose feels hard to sustain (imagine standing like that for 5 minutes) Cats feel like liquid when you pick them up, and when they walk across your desk they know exactly where everything is and place their feet efficiently and never knock things over. Make it so that at any moment it does the most quiet, efficient and aware thing to get to its next goal. Make it walk over those obstacles without kicking them around.
Your projects are on another level!
Boston Dynamics has nothing on you.
They should fear you...
Been loving this whole thing progress! It is so fun to watch you go through every step and overcome all the challenges. It has also made me think critically and come up with potential ways to improve any future iterations of openDog, specifically surrounding the design of the legs. I know that your version of a robot dog is supposed to be able to move forwards and backward relatively easily, with the front and back legs being mirrored, but I think taking a few notes from real dogs might help. In this case, I'll be comparing it to a golden retriever, as I feel it is a good general comparison (also the first high-performance dog I could think of):
1. The first thing I notice is the height of the openDog. In a golden retriever, the length of the legs in a slightly bent, stationary state, results in the height of the dog (paw to top of the body, not including neck or head) being approximately the same as the length of the dog (front end of body to back end of body, again not including neck or head). This should in theory help with control, and the ability to move around much easier. My thinking surrounding this is that the current version's stride length compared to the body length makes it difficult for it to move as effectively. Just to make sure I'm not crazy, I looked at Boston Dynamic's robot dog, and their body length is very close to the height of the robot, so this idea may hold so truth behind.
2. The second thing I notice is the ratio between the upper (thigh kind of area) leg versus the lower (calve kind of area) leg favours the former. In a golden retriever, the upper legs are shorter than the lower legs. The ratio of upper to lower legs is approximately 1:1.5. I don’t know how this could affect the way that openDog moves, with the different structural design concepts applied, but it might help with openDog's ability to handle moving at greater speeds. Again looking at Boston Dynamic's robot dog, they instead have a 1:1 ratio between the upper and lower legs. Additionally, I also noticed that some dogs also share this characteristic. So I can't concretely say if one is better than the other. But I still hold the belief that the lower leg should at least be as long as the upper leg.
I don't have anything else to add, as my experience in this area is limited, due to limited funds to actually pursue engineering challenges (hopefully just for now), as well as still being in high school, but I feel these suggestions might either help with future iterations of openDog directly or indirectly. Again, amazing content! Keep up the great work!
Our cockapoo recognised it as a dog and started barking like crazy!
This is very weird how much closer it gots from Spot in this video. It makes me hope that robots like that could be a lot less than 70k soon
Well done, Sir!
11:14 James's fence looks like a rendering artifact 😆
I find it the sign of human ingenuity, that a person can 3D print parts, cobble together some gears and make a robo dog, where Boston Dynamics spent millions and millions in R&D. Great Job.
I am curious how long it took to print all of those parts.
i'm really impressed!
Would be interesting to see it try a couple different gaits other than Trot. Could it handle a Canter or even a Gallup?
Sample of different gaits: czcams.com/video/PVvZKcKBTtg/video.html
My thoughts exactly
not even Boston dynamics has done that yet lol
Well done James, very impressive, great stability, as good as BostonD's early versions which must have cost 10's if not 100's of millions with a whole team.No more dogs. What next, two legs ?
Stair climbers, I like the snake format, or beatles, they are relativley successful, push legs at the back, balance legs in the middle, climbing legs at the front. Whatever you choose it will no doubt be very well thought out and executed. Like the idea of compliant drives.
Never thought I'd see a robot walk over obstacles like most humans walk over legos
This was fun to watch.
OD3 resembles a baby crawling. Amazing stuff!
Very impressive.
I can’t wait to see how this technology advances. My gf wants a dog in the future and I personally don’t like dogs, but with something like this and some AI to make it kinda just a Smart roomba I think that could be really cool
That is very awesome!
i dont think i could assemble this with my little experience, but it makes me want to try to design my own, more simple robot, and slowly work up to making something like this
We can see the different colored socks James 🧦 🤘
Excellent! I would be interested to see a digitgrade back leg version like a natural dog. You can still use the same number of motors, just use a connecting rod from thigh to ankle to move the foot.
Yeah, it's always bothered me that nobody ever copies the obviously more successful 3-segment hind leg design, yet most biped robots use 3-segment legs when humans do fine with two (birds do use 3, but mostly for higher ground clearance for wings). Though I suppose it wouldn't really make that much difference since they're mainly better for running, and most robots can't run due to excessive weight and/or inadequate software.
You will have to refine your feedback loops but it's almost perfect! For the stairs you now need a Lidar....
I think that it might have been a good plan to first program the inverse kinematics to support the robot as it moves via inertia. The legs should behave like a multidirectional wheel not adding any extra movement and letting it move with inertia. Basically only lifting the leg when its nearing the max range of motion of the leg and then putting it down on the opposite side with a distance from middle point under the leg to the edge of motion which depends on its velocity. The faster it moves to its maximum allowed velocity, the closer to the maximum range of motion in the moving direction should the feet be placed
Thats a 3D printed 2Dog Robot :D
It reminds me of a very old dog, with the way it walks :)
This is an incredible amount of progress. I'm really impressed with how far you've gotten the robot to roll with a very straightforward control scheme. Have you explored implementing a gait engine? Are you trying to avoid adding software complexity or just trying to solve problems at the simplest implementation possible?
Looks really good now! Please make some different gaits for this beast. I reckon a cat-like one leg at a time crawl would work great, and a canter would be worth a shot.
Very impressive, especially as it's largely 3D printed. Don't get me wrong, as a technical exercise it shows how he can can come up with some brilliant and very complex mechanisms, and apply them to a problem. Just because he can, and enjoys doing it. But that complexity is also it's biggest weakness. 12 big cycloidal motors, 6 controllers and a hundred bearings for each motor? But then he glued the silicon feet directly on to expensive carbon fibre tubes, rather than using a jubilee clip or a simple screw. And his "belt tensioner" was a fixed wheel that would have to be replaced as the belt stretched. Simple, cheap, spring loaded tensioners have existed since at least Victorian times. As the great engineerinv genius Colin Chapman used to say, "simplify and add lightness".
Awesome.
Superb!
very cool!
awesome!
I subbed to watch the progress of OpenDog v1. Sad that there won’t be a v4. It’s come a long way, though!
There might be, but just not very soon
I love this!
James: kicks the robo dog
_OpenDog will remember this_
It's never not going to be funny to watch these robots try to walk through obstacles and lightly kick stuff around
This would be perfect for an AT-AT Walker. That would be an awesome build.
So... Freaking... Awesome!!
Could possibly use a small lidar on the front and rear of the dog to enable it to use some kind of primitive path finding to allow it to know when to take larger steps so it's not tripping over stuff.
You could tune out oscillation by using logic to compare gyro out to control input and nullifying the difference.
Awesome! a control system base on AI which learn and adapt feels a natural way to go to make the movement of these kind of robots more stable! maybe!
The dog is back
Soo cool, I wish ODrives weren't soo expensive, makes a big difference between a £200 actuator and a £20 stepper for accessibility for most hobbyists
I’ve been waiting for this haha
I think you can solve the rotation at slow step rates with kinematic cross-coupling of the motor torque. I.e. if the leg position and motor torque is pushing against the ground with 100N, then add enough torque to the shoulder joint to balance that 100N on the ground. Put another way, you have enough info in your robot to make a simple inverse model that takes leg position and robot weight and outputs an "ideal" joint torque. Add this into the output of the PID and then it will be doing less command tracking and more disturbance rejection.
Thanks James. What about something involving its weight carrying abilities? You did a load test on a leg so it would be interesting to see how it spans out with 4 legs.
Part of the instability is likely coming from the fact the “spine” is inflexible. Even just a single motor allowing the front and back to rotate relative to each other might help drastically. Although that would almost certainly require a fully new version so it isn’t a simple solution.
Me *thinking quietly*: "Go home OpenDog, you're drunk."
James: "It looks a bit like it's drunk."
Well now I don't feel so bad about thinking that...haha. Great work as always James.
Комментарий в поддержку канала и ролика, а также труда мастера. И удачи с творческим успехом.
i'm a firm believer that dynamic robots of the future won't have control systems that are programmed and tuned... but rather will have neural networks where it is trained to walk (or whatever else dynamic motions it needs to do). Obviously this is much easier to train if you have a high-fidelity simulation and can run millions of runs to tune the network. But it would be really cool to see if you could do a project where just the raw inputs of the sensors go into a neural network with multiple layers and it is given a goal and allowed to try to learn its own algorithms for accomplishing each goal (perhaps with multiple factors being weighted such as time taken, power used, minimizing joint rates, etc.).
No open dog v4? The open dogs are my favorite content!
Bruton dynamics
Fckn good! Great Job!
not sure if it would be practical but a neural network balancing system would be quite cool for the dog and easier to make as you dont have many inputs other than the sensors though you could add few more to help it. I bet boston dynamics is using something like that
If I understand it correctly the problem at hand is that (in the beginning of the video without the dampening) all the legs always extend all the way.
So in order for the robot to be more stable it would need a way to tell if the leg touches the ground and not extend the leg any further.
Basically you would need to add "feeling" to the feet.
And if I am not mistaken you could add that feeling by measuring the current of the motors. As soon as the leg touches the ground and therefore puts mass on the leg the current of the motor should spike.
You have the relations T = k * I and F = T/(L*sin x) but I think the problem becomes hyper-static.
With four legs on the ground you would need 3 forces and 3 torques for each point of contact. At that point you need more data to solve for the forces.
If you could hook up some strain gauges to the legs you can solve for internal stresses and external forces, so yeah, force sensors pretty much required IMO.