We’ve been following [James Bruton]’s open dog project for a little while now, and with his considerable pace of work – pandemic or no pandemic – development has been incredibly rapid. The latest milestone is the public release of version 3 (Video, embedded below.) This upgrade to the system adds 3D printed cycloidal gearboxes, removing the previous belt drives. [James] had immense fun tuning the motor controller parameters for these and admits they’re not completely dialed in yet. He notes that the wider gearbox body means that the robots geometry needed to change a little, and the previous belt-drive version may have a bit of an edge, but he’s confident he can make it work (and given his incredible previous robotics builds, we totally believe he’ll nail it!)
Older versions struggled with slippery plastic feet; the advantage of a predictably smooth contact shape of a rounded foot is somewhat offset by the limited contact patch size, and that means not so much grip on some surfaces. [James] solution was obvious enough – just learn how to make 3D printed silicone moulds and cast a nice rubber foot around a plastic former, and problem solved! Unfortunately he neglected to add some recesses for a lever to get in between the mould halves, so it was a bit of a struggle to separate after curing. A beginner’s mistake that won’t be repeated, we’re sure.
Full source for openDogV3 is now available on the GitHub page. Here’s the playlist for the whole project, as well as direct links for the cycloidal drive development (part1, part2, part3.) But before you all go diving in to start 3D printing your own pooch, [James] tells us that the total cost would be around $2000 all in, with the bulk of that being the motors and ODrive units, so this one for the serious builder only!
We’ve covered robot dogs a fair bit, a particularly nice example is The Dizzy Wolf, and if you’re wondering just why on earth you’d want a robot dog, then Ask Hackaday has you covered as well.
One of the challenges of many walking robot designs is the fact that they draw current just to stay upright. This was exactly the case for one [James Bruton]’s quadruped robots, where the knee motors were getting too hot to touch. Adding springs to take some of the load is not as simple it might seem, so [James] created a bungee assisted cam mechanism to do the job.
For a normal spring-loaded lever, force is proportional to how much the spring is stretched, which will require the actuators to draw more and more current as it lifts the leg higher. For the spring force to remain constant throughout the range of motion, the length of the lever arm must become continuously shorter as the knee is bent. [James] did this by stretching a bungee cord around a cam. The added bulk of the cam does however cause the knees to knock into each other in some scenarios, but [James] plans to adjust the robot’s gait to avoid this. He didn’t get around to actually measuring the current draw reduction, but the motor temperature has dropped significantly, only being slightly warm after a test run.
By pretty much any metric you care to use, 2020 has been an unforgettable year. Usually that would be a positive thing, but this time around it’s a bit more complicated. The global pandemic, unprecedented in modern times, impacted the way we work, learn, and gather. Some will look back on their time in lockdown as productive, if a bit lonely. Other’s have had their entire way of life uprooted, with no indication as to when or if things will ever return to normal. Whatever “normal” is at this point.
But even in the face of such adversity, there have been bright spots for our community. With traditional gatherings out of the question, many long-running tech conferences moved over to a virtual format that allowed a larger and more diverse array of presenters and attendees than would have been possible in the past. We also saw hackers and makers all over the planet devote their skills and tools to the production of personal protective equipment (PPE). In a turn of events few could have predicted, the 2020 COVID-19 pandemic helped demonstrate the validity of hyperlocal manufacturing in a way that’s never happened before.
For better or for worse, most of us will associate 2020 with COVID-19 for the rest of our lives. Really, how could we not? But over these last twelve months we’ve borne witness to plenty of stories that are just as deserving of a spot in our collective memories. As we approach the twilight hours of this most ponderous year, let’s take a look back at some of the most interesting themes that touched our little corner of the tech world this year.
Walking robots that move smoothly are tricky to build and usually involve some sort of compliant leg mechanism — a robot limb that can rebound like natural physiology for much better movement than what a stiff machine can accomplish. In his everlasting quest to build a real working robot dog, [James Bruton] is working on an affordable and accessible Mini Robot Dog, starting with the compliant leg mechanism.
The 3D printed leg mechanism has two joints (hip and knee), with an RC servo to drive each. To make the joints compliant, both are spring-loaded to absorb external forces, and the deflection is sensed by a hall effect sensor with moving magnets on each side. Using the inputs from the hall effect sensor, the servo can follow the deflection and return to its original position smoothly after the force dissipates. This is a simple technique but it shows a lot of promise. See the video after the break.
A project can sometimes develop a life of its own, or in the case of [James]’s OpenDog, spawn experimentally evolving offspring. This is number four, and it’s designed to be a platform for learning how to make a quadruped walk properly, and to be simple and cheap enough for others to build. We’re looking forward to seeing how it turns out.
[James Bruton] OpenDog remains one of the most impressive home-built robotics projects we’ve seen here on Hackaday, and it’s a gift that just keeps on giving. This time he’s working on adding force sensing capabilities to OpenDog’s legs to allow for more dynamic movement control.
The actuators in the legs are three-phase outrunner motors that drive ball-screws via a belt. This configuration is non-backdrivable, meaning the legs cannot be moved when an external force is, which could lead to mechanical failures. He as tested other backdrivable leg configurations with other robots, but did not want to rebuild OpenDog completely. The solution [James] went with is a redesigned foot with an inbuilt switch, to confirm that the foot is touching the ground, and a load cell attached in the middle of the bottom leg segment. The load cell is bolted rigidly onto the leg segment, which allows it to sense when the leg is carrying load, without damaging the load cell itself.
Unfortunately all the serial ports on OpenDog’s main Teensy 3.6 controller are already used, so he converted the signal from the load cell to PWM, to allow it to be read by a normal GPIO pin. This works well in isolation, but when [James] switches on the motors, the PWM signal from the load sensor gets flooded by interference, making it unreadable. To solve this problem, he wants to implement a CAN bus, which will allow for more inputs and outputs and hopefully solve the interference problem. However, [James] has no experience with the CAN protocol, so learning to use it is going to be a project on its own.
OpenDog is turning into a very lengthy, time-consuming project, [James] says that the lessons learned from it have been invaluable for a number of other projects. This is something to keep in mind with everything we tackle. Choose projects were the experience gained and/or relationships developed are worth it on their own, even when the project fails in a conventional sense. This way you can never really lose.