> You can buy most of the parts directly from a reseller like Amazon or McMaster-Carr, but for some you’ll need to get them custom manufactured for you. The custom parts include the carbon fiber routed parts, the 3D printed parts, the power distribution printed circuit board, and the motors.
Very cool and I would like to build this, but the "Nano Dog" is probably a cheaper and more approachable option for the garage engineer.
> the "Nano Dog" is probably a cheaper and more approachable option for the garage engineer.
Hard to find out much about that or OpenDog in general. I've found videos and 3d printer files but nothing that I can read to understand what it can do and why I might want to build it.
Or do I really have to watch a bunch of YouTube videos? I hate video.
The code's a bit of a mess -- James is clearly not a software engineer first. This is totally understandable, given that he does so many other things!
When I first saw the code, I thought about doing a refactoring pass and making a pull request. It would be painful to test correctness without the hardware, and I don't want to make things worse, so I haven't.
It's probably also worth mentioning the project is a work in progress, and from the videos, it looks like OpenDog is likely to undergo at least one more major change before "completion".
For future reference, I saw "OpenDog" and thought we were discussing James Bruton's (XRobots) OpenDog/Mini-Dog, which is where the firmware I linked above comes from.
NanoDog is something different, apparently! Sorry for any confusion I might've caused there.
Hell, I'm getting flashbacks to my own code in EE lab. Except this stuff looks way less messy, not nearly enough random assembly declarations to bludgeon a register or pin into submission.
Good point, now that I am searching for it I cannot seem to find the code. At the same time it still does not require custom manufactured parts or circuit boards which makes the barrier to entry (and importantly price) a bit lower unless the Stanford folks setup a store.
I make an internet promise here and now that if I ever build the robot I'll open source the source code to control it.
For carbon fiber parts (used for the chassis and legs), I wonder if it is possible to carve them out out from the same power distribution PCB board. Some PCB manufacture seems to support custom shapes.
But the PCB material must be heavier than carbon fiber, not sure if it will affect the robot dynamics.
Just for fun I uploaded the top and bottom body panel DXFs to jlcpcb and they're only about $2 each. That's compared to about $30 to have them cut out of carbon fiber, and of course you don't get to put circuity like your power distribution on it for free.
I'd be amazed if you found any doing a dynamic gait with closed loop control. One problem with making legged robots smaller is that you drastically increase the reaction speed required for any kind of closed loop control.
I've built several servo powered walking robots. Torque is a major concern, and obviously weight plays a part in how much torque is required. The recommended servos look to be $30 each, x12 means $360. If you spent $200 just on servos you'd have to settle for $16 servos. Probably less torque. That would mean that at full leg extension the robot might not be able to support itself. That would be an annoyance when trying to focus on walking algorithm development, which is usually the fun part for these small walkers.
You can definitely do something for cheaper, but I wouldn't call this a joke. When I was 15 (20 years ago lol) I designed an 18 servo hexapod. I saved up $350 from cleaning horse stalls and my dad agreed to pay the other half, and I spent $700 on servos. It was great to learn about inverse kinematics before ever setting foot on a college campus!
If the 35kg is in fact needed (which I’m not prepared to debate, probably needed, but I can’t prove / disprove), you are exactly right. Yes, you could maybe save a couple dollars with cheaper servo brands, but honestly for a 35kg servo, you’re going to pay a decent amount for 12 of them.
This gihub link may be a better one, as it provides demos, hardware, software, and design descriptions along with the schematics, pictures, and ICRA paper reference: https://github.com/Nate711/StanfordDoggoProject
Awesome project Nathan, kudos to the team. I'd love to use the Stanford Pupper at some point for K-12 outreach efforts myself.
I reviewed your power distribution board. I think it needs more safety features before it can safely be entrusted to the hands of K-12 kids. Couple of comments:
- The battery input is unfused. I would HIGHLY recommend adding one.
- I know you guys wrote in Github about how people should not plug in voltages above 8.4V, and not overdischarge batteries, but I would not trust K-12 kids to follow those instructions :)
It would be best to add a overvoltage, overcurrent and reverse polarity protection circuits. Those are pretty simple and should not take much space on the board. Jerri Ellisworth has a great video where she explains these circuits in detail, and how to make them low-cost:
https://www.youtube.com/watch?v=QZ0WXQWND-I
- I would add a schmitt trigger circuit to sense the battery voltage and feed that as a digital input to the Raspberry Pi. If the batteries are too low, the Pi then can throw an audible alarm and disable the servos. This would prevent kids from overdischarging their batteries.
As a future addition, I think it would be sweet to add a Teensy somewhere. Then students would be able to easily add extra sensors and have their robot respond to changes in the environment etc. So much potential for fun.
Awesome project! I'd love to build one, except for sourcing the parts - esp. the carbon fibre (out-of-stock) bits! If your team sold it as a kit I'd be your first customer! Of course, you'd have to ship to Canada too.
Q. Was there a price point you had to aim for in the development (i.e. under $1K)? And, if you were to double the price, what improvements would you make (e.g. even better servos)?
Hi Canada_dry, We are selling Pupper kits now. Please check out our website link below If you have any questions you can can contact us at:cypresssoftware@gmail.com
We're actually working with some people who are interested in selling the kits as a side project. Stay tuned. With double the BOM cost, we'd use torque-controllable servos. We're testing out the DJI M2006 right now.
Martin's robot was definitely an inspiration. As far as I know he was the first one to figure out how to make cheap, nonbackdrivable servo motors work for dynamic quadruped movement. Wouldn't have started the project without seeing his videos.
It's fun seeing these. As batteries and motors get better, they get more agile.
I'm disappointed that there's no force feedback. They're still using dumb PWM servos, which is a 1970s technology. There are better servos, from Dynamixel, where you get info back. There was a project called OpenServo [1] to smarten up dumb servos by developing tiny new controller boards for cheap RC servos. But that project seems to be defunct.
There are lots of good legged toy robots in the $300-$400 range, but without good feedback you can't do Boston Dynamics type stuff.
We've found you can actually do a lot with dumb servos and orientation feedback and get visually similar results to Spot. For instance, we get hopping, trotting, strafing, turning + strafing, walking. What it can't do is prancing or a flying trot. But we think it's totally worth it since a good-enough servo is around $20, while similarly sized torque-controllable servo (excluding dynamixels, which don't actually have torque control), runs about $400 each. DJI did actually come out recently with their M3508 and M2006 which we're starting to play with, and we think we can build a torque-controllable pupper for about $1.5k.
The main limitation I think for servo robots with kinematic (position-control) controllers is speed. We're moving around 0.8m/s max, while the new Unitree A1 is something like 3.5m/s.
From watching the demo video: This thing seems amazing and for around USD 600, buildable in 8 to 10 hours, within hobbyists reach. The demo reminded me much of the Boston Dynamics ones but without the creep. I guess that has much to do with the fact that the Pupper is much smaller than Boston Dynamics' monsters.
EDIT: They latter address this in the demo video: "We specifically avoided to make them just look like dogs. I think when you're on[sic] the uncanny valley that would be a little bit weird."
I came across this project while working on my newsletter and thought HN might appreciate this. With the $500-600 price I'm heavily thinking of building one for myself.
If someone would like to build something larger then Stanford Dogg (https://github.com/Nate711/StanfordDoggoProject) is one choice, however as far as I remember the price comes to around $3k for parts only.
Even cheaper alternative to Pupper could be OpenCat (https://www.petoi.com/) however it was never fully open source and I don't think it's available anywhere.
This is the kind of the stuff that makes me wanna dust off my raspberry. I do have a vacation coming up. Maybe I should pencil in some fun time since it is a staycation anyway.
I know it did not sound right, but between forced stay at home, in practical sense, we can't really go anywhere so we agreed on some home projects. It is not a true vacation, but it will be departure from regular routine.
For about the same price, you can build a "spot micro" [1] if you have a 3D printer. I haven't build one myself (yet) but it's been fun following them. I came across them when someone in HN posted how to train it using OpenGym (will find the link when I'm not on mobile).
Watching this is really heartwarming. It's inspiring to see young talented minds directed towards curiosity and engineering while also sharing the fruits of their labor for everyone else to benefit. Kudos to the team!
I don’t want to talk bad about anyone’s project. This is the first time I’m seeing both of these projects and the Stanford Pupper looks a lot more promising.
My heuristic is the following: Legged robotics is hard. Four legged robots are even harder than six legged ones. Using hobby servos makes the task even harder. You can find a lot of succesfull hexapod projects with hobby servos but up until fairly recently no four legged ones.
These robots operate in a very thight spot of the design space. The motors have to be fast enough to be responsive, but torquey enough to bear the dynamic loads. The mechanism has to be compliant enough to not overstress the gears but not too springy lets the robot collapse into a mess. The structure has to be rigid enough to support the forces, but also light enough so the legs can carry it. All in all designing a four legged walking robot with hobby servos is a hard problem. It is very easy to put together something which looks like a dog robot and even easier to feel that once the hardware is there from then on it is only a software problem to make it walk. For many possible designs that is false. Not even the best software could make them work.
This is what i know about the problem space. And then I look at the spot micro. It looks awesome. Clearly they made it to mime the spot mini robot. The problem is that all that is superflous. It is a design fetish. Maybe they put in all the effort to design the right leg lengths and the appropriate springyness and carefully considered the torque and thermal limits of those servos and come to the conclusion that they can afford the extra grams. But it is more likely that they did none of that and just made the 3d drawings look cool.
If you build the Stanford Pupper you will eventually probably end up with a walking robot. The video where they show it perform on the stage is a pinacle of a lot of tweaking and design work. I just don’t see the same about the spot micro.
The SpotMicroAI project has a great community on its Slack, but the project is a bit more in its infancy than Pupper. For instance, spot micro doesn't have the software right now to make it walk quickly or trot. Pupper is at a v1.0 stage with software and build instructions already complete, but it's potentially more expensive to build than spot micro.
I want this but where it's self-piloting/long lasting(battery) and can build a point-cloud of my apartment as it navigates. Also one for outdoors just wanders in a farm or something... I don't know has no purpose but "neat". Maybe you program it to do a routine where it walks around takes photos of things throughout the day, self charges.
edit: to clarify this isn't a negative comment/criticism. I'm saying I fantasize about having a platform like this but with the capabilities I mentioned(which I'm aware is not easy).
Currently I can't even program/make that gait that makes it move so smoothly so I'm definitely not criticizing.
What is the cheapest robot you can build? This robot is pretty expensive, is much more than a Lego Mindstorms EV3, Bandai Smartpet Robot Dog is USD 157 in Amazon.
I am mainly thinking about a "one robot per child" initiative.
There seem to be two versions of the Bandai Smartpet Robot Dog on Amazon, one in black for $157 (that's out of stock at the moment) and one in white for $60. I don't know why there's such a high price difference, especially since there doesn't seem to be any functional difference between the two.
Gotta say, really comforting that this was written in python. Everyone I talked to when it comes to quadrapeds tells me python is too slow for my projects.
People always imagine that mobile robotics needs nanosecond response times or some other silliness when the reality is that your sensors and servos only operate on the order of 100Hz.
You can buy the CLS6336HV servos from JX-Servo directly for about $25 each. Usually a much better deal than on Aliexpress, Banggood etc and it's pretty easy to do. The $600 price point though assumes you have a number of things already like a RPi, lipo charger, or PS4 controller.
I summed the BOM table it was just over $330 but it looks like that's just body no electronics at all... the servo arms look special/priciest but it's multiplied. I imagine the main thing for this aside from body is the brain but that looks to be shared as well
Very cool and I would like to build this, but the "Nano Dog" is probably a cheaper and more approachable option for the garage engineer.
https://www.thingiverse.com/thing:3145690