ROS controlled Bioloid with custom sensors

[m3atsauc3]

Update: I re-generated the self-collision matrix, and now setJointValueTarget() does not end up registering collisions which is good! I was possibly either trying some impossible move previously, or the re-generation of the collision matrix solved the issue, either way I'm happy! Now to figure out why the trajectory generation method still refuses to work...

[LloydF]

wow, that's amazing you are farther along than the HROD1 Folks. Or there not working on it at all.

[m3atsauc3]

Thanks! I don't have any idea how far along HR-OS1 is, I've been meaning to have a look into it for a while now.
Lately I've been working on tidying up the hardware side, and this week I got my very first delivery from Shapeways of a custom 3D printed backpack for the Bioloid! Although a bit pricey, the print accuracy is much better than I could hope from my old RepRap! I just need to sort out the foot sensors and the hardware will be largely done (for now).



More info here:
https://dxydas.wordpress.com/2015/08...kpack-printed/

[m3atsauc3]

Small hardware update: The robot now has a frame for use while working/testing, which makes life so much easier than trying to find a balanced resting position with the robot standing upright!
And back to software: Progress is slow, but I've been working on a GUI which will help me test out ROS/MoveIt! libraries much easier than using small rushed test programs.
Hopefully it won't be long before I can show a video of the Bioloid doing something remotely interesting!

Updates as usual over here.

[m3atsauc3]

The control GUI has been updated with a number of new functions now, however I have had more important progress with integration into MoveIt.
It took me a while to get my head around how the controllers are supposed to be loaded and integrated with MoveIt, but I finally managed it. Here are the two most important links out there for anyone interested: controller configuration and hardware interface.
So, the loop is now closed and MoveIt trajectories can be executed on the real Bioloid!
Unfortunately the default kinematics solver is struggling to creating efficient motion plans, even for the simplest moves, so some more work is needed on understanding exactly what the problem is. It is quite exciting however to see the real robot actually perform some complex arm and leg movements, using the virtual robot in RViz.

I know I'm repeating myself from last post, but hopefully I'll have some interesting videos to show soon, instead of just boring text!

[Mathieu]

I got inspired by your work so I hope this is okay, I forked your project to be able to re-use the cad files and I have integrated it with the HumaRobotics Darwin-OP projects to get the walking engine going:

https://youtu.be/kJ-ZhCndFq0

I had to change the orientation of parts and joints and add the gazebo xml tags required (inertia stuff). I put the changes in github:
https://github.com/MathieuR/bioloid_typea_description

you would also need the following two repos to get it to walk in gazebo:
https://github.com/MathieuR/bioloid_typea_control
https://github.com/MathieuR/bioloid_typea_gazebo

I did not invented anything, just integrated two projects together so credits goes to you and HumaRobotics guys.

I hope this is okay. I learned a lot!
It was my first time working with ROS and Gazebo.
It was lots of fun.
I am now looking at trying on a real robot

-Mathieu

[m3atsauc3]

Glad it was of use! I'll check out your video later today.

The Gazebo inertia params were only placeholders so I could get the robot into Gazebo as a quick test. A good source of experimental inertia values I've found is this.
Also, the new version of Gazebo has a graphical editor for URDF files, which sounds extremely useful although I haven't tried it yet.
Source

[Mathieu]

Thanks for the links, I really appreciate it, the inertia values is exactly what I was looking for.
I took the ones from DarwinOP for testing (it seems ok) and I also try to use meshlab but I gave up!
I will update the xacro files with those values.
Thank you again!

[m3atsauc3]

Hi all,

I realised it's been quite a while since I updated this thread, so thought I'd make some notes on the progress. I've been working on a few various things, but as always I wish that they would progress faster than they have been!

• The robot hardware how now been put together, including 6 foot sensors (ideally I will increase them to 8 to cover all four corners).
• The custom GUI has progressed quite a bit, with a variety of functions now added, such as a PID controller UI, individual motor interfaces and sensor data plotting.
• On ROS, one of the challenges was getting to understand how the various ROS "controllers" and packages operate, and specifically how they integrate with MoveIt. I finally found that I had to implement the "hardware_interface" for my AX-12+ motor controller, and also figured out how to configure the controller YAML files and actually get the "controller_manager" to spawn the controllers correctly. Creating custom roslaunch files as well as notes on how to run each one helped a lot, as it can get quite confusing!
• I've been learning more about MoveIt, and found that the easiest way to set up pre-defined poses (as you would do say with the stock Bioloid software) is actually throught the MoveIt setup assistant. Then, the RViz plugin can be used to perform the path planning between poses, although I have found that this crashes often for some reason I haven't found out yet.

I'm currently not sure where I should focus my attention on more, so was thinking of asking you guys/girls what you think would be a good idea. Perhaps a demo of some sort? The humanoid form is quite a challenge, as e.g. the arms themselves have limited range of movement and no hands. Ideally I would like to get the robot walking while using the orientation sensor information for balance, but this alone is a major undertaking!

Below are some pics of recent progress:

ROS:


GUI:



Robot:

[jwatte]

Ideally I would like to get the robot walking while using the orientation sensor information for balance, but this alone is a major undertaking!

Assuming that you have functions to make each extremity reach a point using IK, that sounds like it would be the next logical step!