Xl-320 Based Hexapod

[Shotch]

Hey all, just putting this thread up to kinda log and show off my progress on my xl-320 based hexapod.

The biggest limiter on this project so far has been dealing with cost, working on stuff like this with a highschooler's budget has been hard, but I hope to work on many more projects in the future with better hardware once i'm able to afford such.

Anyways, the brain of this project is just a raspberry pi 3, sensors include a lidar lite and an ADXL345.




As is probably very very evident from that photo, it has some bad load issues. Some back of the napkin calculations led to some pretty wicked part sizing issues. But, given my limited capacity to print at the time of initial printing of the frame, I proceeded with construction anyways.

Parts were designed and modeled in OnShape, parts are variable sizes with variables in the construction so alterations are easy.


After doing load testing and initial kinematics work, the model has been refined with what I assume now to be much more reasonable numbers for loads.



In order to test IK and FK code before actually having a robot build or modeled, I started working on a simple sim program so I could simply migrate my IK and FK classes over to the code eventually deployed on the bot itself. It can also stream out and take in inputs for the xl-320's through the pi over wifi.




Current plans are to continue printing out the new frame, print out rubberized 'socks' for the legs, and continue making revisions on the framework, and optimize the code a bit more in order to try and resolve jitter (could probably tune PID on servos a bit better) and make further development easier.

So far I've done full FK tests for rotation and translation, and the code is in place for walking, which is can do while suspended, but currently can't handle the load unsupported. (maybe wave or ripple gait?)

Anyways, here's FK test:




This project has been my first real foray into robotics, and programming, and 3d modeling, so any tips and such would be greatly appreciated. I maybe should have taken the easy way out a while ago, and tried to use more existing material out there to get this working faster, but I feel I've learned a lot along the way, and am excited to continue working.

[jwatte]

That looks great! And a Raspberry Pi 3 is not a "just" -- it's one of the best lightweight robotics controller boards on the market!
To manage the weight problem, keep pushing those knees inwards to get a shorter lever arm, and perhaps use a gait that puts five feet on the floor at all times?

[Shotch]

That looks great! And a Raspberry Pi 3 is not a "just" -- it's one of the best lightweight robotics controller boards on the market!
To manage the weight problem, keep pushing those knees inwards to get a shorter lever arm, and perhaps use a gait that puts five feet on the floor at all times?

Yeah, one thing I'm working on code wise is to read the load from the servos and use that to dynamically adjust the 'rest' feet positions in order to minmize that distance. Just to keep it under a certain threshold if possible. I feel that the new frame will really help out though but that remains to be seen, but I'm eager to test once I finish printing it.

I may actually leverage my gait generator so I can see what happens if I only make some legs shorter, just to further test IK stuff as parts come off and I rebuild the legs as is.

[Brooks]

Very nice to see - and impressive! Keep the photos coming!

[KevinO]

This is great! Well done!

[r3n33]

Love the look of this one.. Nice stuff!

[Shotch]

Hey all,

So, i've made a good deal of progress over the past few weeks.

I went through and planned and then implemented my curved path walking system. It essentially offsets a point of rotation for the entire robot, inversely proportional to the x axis input. With that pivot point an arbitrarily large distance away, the circumference of each circle formed by the legs are essentially straight lines, leading it to continue to walk straight, though as the pivot point gets closer to the robot, it causes the legs closest to the point to correspond to a smaller circumference than the larger legs. This allows it to walk around specific points in space, as opposed to purely stand in place rotations, and then transnational motion.



As that was going on, I finished printing the new frame, except for the hub, which ran into some issues, in which I'll need to do almost a full redesign to allow for more electronics space and rigidity. I thought the scale on the shorter legs would make the body look out of proportion, but I kinda like it, and with the eventual frame redesign, it should come more back into line.



I did a real big push over the span of roughly two days to finally get the thing untethered. Setting up a static ip and wifi connections, getting remote deploy working on my IDE, dealing with the pi3's wonderful wifi config.. ect ect. Making room within the current frame to route the electronics by splicing new longer cables from the legs. Getting automatic ps3 controller identification, connection, and trusting over Bluetooth so I would no longer need an initial wired connection to pair. And then lastly, getting the pi3 local power. Now, local powered turned out to be a bit of an issue given the lack of available space, and me waiting on new T plug connectors to splice power off the lipo into a small voltage regulator for it. So I threw on a small portable battery bank, and ran power from there.

And lastly, all that culminated in the first untethered gait test, assuming the servos can hold up to angular acceleration asked of them, The speed in the clip is only roughly 40% top speed with the current gait.



I'd like to thank ya'll for your kind words and it's really been pushing me over the past few weeks to do better and better on this.

Next goals are to refine the software backend, and implement the lidar lite scanning for eventual autonomous navigation. I hope to use a small imu, with the body FK in order to take semi 3d point cloud maps for enhanced navigation in the future by rotating the plane that device can scan in with the help of another xl-320. The data will be sent to a NVIDIA Jetson Tx1 ground station for processing and motion path generation.

[Gertlex]

Awesome work, and great documentation so far! The bot is very clean looking!

Looking forward to more.

[Shotch]

V4:


It turns out, that this isn't probably going to be made for a while, this is the V4 version of the XL-320 based heaxpod lineup I've been working on. Same electronics as the last one, new frame, better structure, actual planning involved.

Software wise a lot has happened, working on trying to make it easier to use and expand to future legged robotics projects.
Added in an implementation of the FABRIK Ik engine, so arbitrary DOF legs could be used, as well as full body kinematics to drive the leg end positions if needed.
Slowly working on my gait engine to support the enabling / disabling of legs on the fly to use them as manipulators instead of for purely locomotion. Getting those transitions down without going to a complete stop is rough.

Plus side of nice modular software that's driven only by config files for the specifics of the robots, I can work on many more 'modules' for the limbs and also, other robots entirely. This leads me into the biggest development as of recent. I managed to get my hands on some AX-12a's, so I put actually building V4 on the backburner and went full steam ahead on a new AX-12 version.

V2-1:

Plans are to eventually swap out the lidar light on the front for a 2DOF binocular camera setup and mount a proper spinning lidar setup on the back plane.

Almost all the structure is 3d printed and so far the total is over 250 hours and has encompassed a lot of my freetime over the past month, but, it is so far complete.

[Note: Belly plate and front cap not printed at the time that image was taken]

At this point just waiting for connectors and pins to come in to work on the wiring harness for it, software should be ready to just plug and play with a new config file and I hope that the margins I have to work with on this one are much higher than the V3 XL-320 based one.

Sorry for the lack of updates, but life tends to happen and makes the timescales of these sorts of projects draw out.

[jwatte]

Still looks great :-D