[Question] The available robots models are not working with the DQ_QuadraticProgrammingSolver

[marcos-pereira]

Bug description

• The available robots models are not working with the DQ_QuadraticProgrammingSolver.

To Reproduce

Code

from dqrobotics import *
from dqrobotics.solvers import DQ_QuadraticProgrammingSolver
import dqrobotics.robot_control as dq_controllers
from dqrobotics.robot_control import ControlObjective
from dqrobotics.robots           import KukaLw4Robot

robot = KukaLw4Robot.kinematics()
qp_solver = DQ_QuadraticProgrammingSolver
controller = dq_controllers.DQ_ClassicQPController(robot, qp_solver)
controller.set_control_objective(ControlObjective.Pose)

Output

Traceback (most recent call last):
  File "qp_solver_issue.py", line 9, in <module>
    controller = dq_controllers.DQ_ClassicQPController(robot, qp_solver)
TypeError: __init__(): incompatible constructor arguments. The following argument types are supported:
    1. dqrobotics._dqrobotics._robot_control.DQ_ClassicQPController(arg0: dqrobotics._dqrobotics._robot_modeling.DQ_Kinematics, arg1: dqrobotics._dqrobotics._solvers.DQ_QuadraticProgrammingSolver)

Invoked with: <dqrobotics._dqrobotics._robot_modeling.DQ_SerialManipulatorDH object at 0x7f3b34cfa3f0>, <class 'dqrobotics._dqrobotics._solvers.DQ_QuadraticProgrammingSolver'>

Expected behavior

• The DQ_QuadraticProgrammingSolver should be created without errors.

Environment:

• OS: Ubuntu 20.04
• dqrobotics version: 20.4.0.27
• Python version 3.8.5

Additional information

• I believe there was some change in the DQ_QuadraticProgrammingSolver that used to accept KukaLw4Robot.kinematics(), but now it requires the newer DQ_SerialManipulatorDH. However, the KukaLw4Robot was not updated.
• If this is the case, it would also be interesting to know the right way to create our own robots objects in Python. I was using the following code to create a robot:

  from dqrobotics import *
  from dqrobotics.robot_modeling import DQ_SerialManipulatorDH
  import numpy as np
  from math import pi
  pi2 = pi/2
  def KukaLBR4pRobot():
  kuka_lbr4p_theta = np.array([0,     0,     0,   0,   0,    0,   0])
  kuka_lbr4p_d = np.array([0.2,     0,     0.4, 0,   0.39, 0,   0])
  kuka_lbr4p_a = np.array([0,     0,     0,   0,   0,    0,   0])
  kuka_lbr4p_alpha = np.array([pi2,   -pi2,  pi2, -pi2, pi2, -pi2, 0])
  kuka_lbr4p_type = np.array([0, 0, 0, 0, 0, 0, 0])
  kuka_DH_matrix = np.array([kuka_lbr4p_theta,
                             kuka_lbr4p_d,
                             kuka_lbr4p_a,
                             kuka_lbr4p_alpha,
                             kuka_lbr4p_type])
  kuka_lbr4p = DQ_SerialManipulatorDH(kuka_DH_matrix, 'standard')
  return kuka_lbr4p

[mmmarinho]

@marcos-pereira

Thanks for the report.

I think this bug is not present when using dqrobotics/cpp#34. However, it might take us some time to update it.

[marcos-pereira]

@mmmarinho , thanks for the answer. In the meantime, I will migrate some of my python codes to cplusplus.

[mmmarinho]

@marcos-pereira

Sorry, but there was actually no bug. The minimal example you sent is incorrect because DQ_QuadraticProgrammingSolver is an abstract class and cannot be instanced (it is not supposed to be).

You need to choose a concrete subclass of DQ_QuadraticProgrammingSolver instead. In Python, I recommend starting with DQ_QuadProgSolver.

from dqrobotics import *
from dqrobotics.solvers import DQ_QuadprogSolver
import dqrobotics.robot_control as dq_controllers
from dqrobotics.robot_control import ControlObjective
from dqrobotics.robots import KukaLw4Robot

robot = KukaLw4Robot.kinematics()
qp_solver = DQ_QuadprogSolver()
controller = dq_controllers.DQ_ClassicQPController(robot, qp_solver)
controller.set_control_objective(ControlObjective.Pose)

To use it, don't forget to install quadprog. python3 -m pip install quadprog --user

[marcos-pereira]

@mmmarinho many thanks! I was forgetting to install quadprog. Maybe you could add this information to the ReadTheDocs Python installation. Similar to the C++ interfaces where you mention the need for CPLEX.

[mmmarinho]

@marcos-pereira

Indeed. In fact, you're welcome to update the docs with that information and generate a pull request.