Quadbot 17 - Quadruped robot designed in Autodesk Fusion 360

[m3atsauc3]

Hi all,

I'd like to introduce Quadbot 17, a quadruped robot born out of a learning exercise in Autodesk Fusion 360. The robot currently only exists as a CAD model, but with any luck it will become a reality soon!

It will use AX-12A Dynamixel servos (might be upgraded to the more powerful AX-18A). The legs currently have 20 DOF and there are an additional 2 DOF for the body. Servos and brackets are from the Robotis range, with some replaced by their metal counterparts available from Trossen Robotics. They will be painted to match the colour theme. The rest of the parts will be mostly 3D printed where appropriate. The main framework of printed parts forming the body will be sandwiched between 1.6 mm thick custom (laser-cut?) aluminium plates.


The novel aspects of this quadruped are its articulated legs, which have higher DOF than usually found on small quadrupeds, and its articulated "spine", which will help it in navigating uneven terrain.


This robot shares many ideas and knowledge gained from my other Dynamixel-based robot, the ROSoloid. Whether or not it gets integrated with ROS and the custom GUI remains to be seen. At the very least, the Raspberry Pi will run the kinematics and walking gait code.


One main thing I felt the ROSoloid humanoid robot was missing was an advanced sensory system. The advantage of the quadruped is its capability to carry heavier loads and provide a stable platform for sensors, hence the idea to add a Kinect! Its size seems to be perfectly fitting with the rest of the profile.


The build progress of this robot can be followed here.

Comments, suggestions and feedback are always welcome!

[Brooks]

Very cool! I'm big on foot sensors - add them in if you can! And I would imagine you'll want metal plates in the legs...

Keep us informed!

[m3atsauc3]

Thanks! Sensors such as the FSRs can be easy to add. I've already strengthened the feet with metal plates, and updated the rear bumper, which now fits a 2200 mAh battery nicely.
The main issue is that this is largely a theoretical exercise for now, as the electronics are around the £1000 mark, and the 3D printing another £1000 (or simply a new 3D printer!).

[RoboChris]

Freaking awsome mate. Did you end up scraping the money together to turn this beauty into life?

Spela på allacasinopanatet.nu idag!

[jwatte]

1.6 mm thick custom (laser-cut?) aluminium plate

I think water jet is much cheaper. Laser cutting aluminum requires ridiculous power because of the high heat conductivity.
(This is one reason why smaller laser shops only cut stainless steel; it has lower heat transfer and thus requires lower power to cut.)

Do you have enough clearance between the two foot servos to get the power/signal wire in there?

Also, I'm not too keep on FSRs because they are fragile and inexact. A hard spring plus an optical or Hall displacement sensor might be better.

Anyway, I, too, want to see this robot built :-)

[m3atsauc3]

Thanks for the advice jwatte!

I'm going to build a test rig for one full leg soon, without decorative 3D parts. Even that is not cheap!

The forward kinematics are calculated, but that was the easy bit. IK shouldn't be that hard, a closed-form solution should be relatively easy as most of the leg moves along a plane. I think treating the foot separately will also make things easy (as a start, foot will always be co-planar with body, then can be adjusted independently for moving on uneven terrain). That's the very rough plan anyway!

[m3atsauc3]

The Inverse Kinematics are now also done, just in time for the hardware.
I am still going with the idea of always keeping the foot perpendicular to the body/ground for now, I may have to make it more complex later.
Hopefully the next update will show this working on a physical test leg!

[Brooks]

Need pictures!

[m3atsauc3]

Hardware parts received! AX-12, Trossen metal brackets and ArbotiX-M controller. Some parts do not have metal counterparts, so I used spare plastic parts from my Bioloid kit.
I have put together a test leg along with a frame made out of Makerbeam aluminium profiles.
Unfortunately, I thought I had more 3-pin cables than I actually did, so I can't control all the motors just yet. However, I've got an Xbox One controller controlling the IK's foot target position, and a simple serial program on the ArbotiX-M which receives the resulting position commands for each motor.
All I need now is the cables, then I'll be able to show a video of it actually working.

[m3atsauc3]

It moves! An Xbox controller joysticks moves the foot target position, and IK is solved such that the foot tries to stay in parallel with ground.

I've tried to get the position updating as fast as possible, although I occasionally lose leg control and I'm not sure yet if it's serial failure or something else. The Python script sends new positions every 50 ms, serial baud rate is 38400.

The servos initially juddered a lot with default AX-12 settings, particularly the ankle and knee. This was improved largely by increasing the compliance slopes. Changing the compliance margins, punch and speed didn't have much effect on the judder.

Unfortunately I didn't realise the video quality was so bad until I played them back, so I will have to try again later!
This is all I have for now: