Code for driving a robotic arm

[oliverks]

Do you have any code you could share for driving the physical arm implementation from your paper. I understand that this would not be general purpose, but it would be helpful to see how you did it.

[costashatz]

The simple_arm example is the closest code we have for the real robotic arm (but in the DART simulator). The main change one would have to do, would be to edit the execute function (see here and here) in order to execute a specified policy on the real robot and collect the needed data (i.e., joint angles and commands for this case).

In our IROS paper, we experimented with our custom dynamixel-based manipulator, omnipointer. To control it, we are using our libdynamixel C++ library which is quite easy and intuitive to use. We also have a ROS control node for it (see here); for examples on how to use it, see here (this one is for a similar manipulator) and here (this one is for our hexapods).

Overall the code for the simple_arm example should be enough for someone to replicate the results of the real robotic arm of our paper. The main change would be to create the execute function that executes a given policy on the actual hardware and returns the joint angles and commands (for the model learning part); this part of the code is very specific to the actual hardware one would like to use and should not affect the other parts of the code.

P.S.: We have in our plans to release a newer, easier to use version of the Black-DROPS code (although the main ideas/functionalities are the same); should be out until end of September..

[costashatz]

I forgot to mention that the best way to edit the execute function would be to create a new Black-DROPS system class. You can take ideas from the ODESystem class (see here); basically you only have to change the execute function..