Add easily available QP solver backend

[rtkg]

To ease sharing the code, it might be nice to integrate an easily available (either directly in ROS or via debian) QP solver as default without the need to go through the hoops of acquiring a Gurobi license. Then, there could be a launch file parameter to switch between solvers if users have, e.g., Gurobi set-up.

Possibilities could be:

• qpOASES
• CGAL
• IPOPT (actually a full-blown non-linear programming solver)

[neckutrek]

Yes this is a good idea. One could use defines in the source code to allow switching backends at compiletime as well. Le 17 déc. 2016 13:51, "Robert Krug" <[email protected]> a écrit :

…

To ease sharing the code, it might be nice to integrate an easily available (either directly in ROS or via debian) QP solver as default without the need to go through the hoops of acquiring a Gurobi license. Then, there could be a launch file parameter to switch between solvers if users have, e.g., Gurobi set-up. Possibilities could be: - qpOASES - CGAL - IPOPT (actually a full-blown non-linear programming solver) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <https://github.com/neckutrek/hiqp/issues/22>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADmjhd8S-GR1afDCJ1D810j-q2jBhR6sks5rI9qwgaJpZM4LP4Tn> .

[rtkg]

Yes, another option. However then a user would need to compile the code herself, which might not be so nice with having a ROS release in mind where the user would just get the binaries.

[ksatyaki]

Hello guys,

I have started work on using qpOASES as the QP solver.
I have:

• made a copy of the GurobiSolver class. Renamed stuff appropriately.
• changed some structs inside the copied class to suit qpOASES port.

Looking at example1.cpp from qpOASES, it seems to me that an approach similar to the final assignment of the Robot Control course should work. Since I am not familiar with gurobi, I want a little bit of help from @rtkg or @neckutrek . Hopefully you should be able to answer this very quickly.

I am not sure how gurobi fixes the slack variables. Please take a look at the fairly simple user manual for qpOASES. I want to know if my understanding is correct. (All symbols are the same as in page 19 of the qpOASES manual).

• For geq tasks, we need to set lbA to the rhs constant and set ubA to +INF.
• For leq tasks, we need to set ubA to the rhs constant and set lbA to -INF.
• Matrix H is always square with dim = (no of joints + no of dims in current stage).
• Matrices H and A will have to be constructed with x = Augmented w-q_dot vector in mind.

So if we are solving the second stage and if first stage had 2 tasks and a second stage has 3 tasks,

A =| J(2 x 1) 0(2 x 3) |
. . .| J(3 x 1) I(3 x 3) |

J is task jacobian
0 is a zero matrix and I is the identity matrix.

Does it look right?
I am guessing that this is what is done in Gurobi too, but gurobi itself takes care of some of the details as I only see Gurobi Linear Expressions in use.

If this is correct then there is very little work to finish to have a qpOASES solver back-end in place. HiQP can then be released !

Thanks
Chittaranjan

EDIT: A version that compiles and successfully executes the joint_configuration task has been uploaded. Needs further debugging and extensive testing though. Let me know what you think.

[ksatyaki]

And by the way, I've opened devel-qpoases for this purpose.

[neckutrek]

Hi I don't remember the math in that detail. Robert lended me a good book on convex optimization, maybe you can find something in it that suits this. Would be cool if it were released! :) //Marcus

…

--- Med vänlig hälsning / Best regards, Marcus A Johansson +46 (0)70 315 73 77

On Sat, Aug 26, 2017 at 12:48 PM, Chittaranjan Srinivas Swaminathan < ***@***.***> wrote: And by the way, I've opened devel-qpoases for this purpose. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#22 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADmjhR_YYGNBCOTP8ZNK45OhHQnRJPy2ks5sb_gZgaJpZM4LP4Tn> .