mahaarbo / casclik Goto Github PK

Currently PseudoInverseController only supports 1-D SetConstraints. This stems from how we generate our in_tangent_cone_func. They're easily defined as 4 if-else's in 1-D, but I can't think of an easy way of doing it in higher dimensions just yet.

If I recall correctly it's just in setup_problem_functions that it is required.

Currently the ModelPredictiveController does not have an initial problem solver. This can be a problem for situations where slack variables are required, or virtual variables. Not as big a problem if you're using IPOPT.

Need some unittests. Minimal tests that can help:

1. Checking jacobians of a simple 1 D equality constraint
2. Create a SkillSpecification and count constraints
3. Setup ReactiveQP, ReactiveNLP, PseudoInverse, and ModelPredictive controllers and run once to see if that works.

PseudoInverseController has support for virtual vars, but doesn't actually try to solve them once so that they fit.

Modify to allow for just a list of floats/ints and convert them to numpy arrays internally?

    in_tc_func = self.get_in_tangent_cone_function(cnstr)
  File "/home/mathia/.local/lib/python2.7/site-packages/casclik/controllers/pseudo_inverse.py", line 146, in get_in_tangent_cone_function
    ["in_tc_"+cnstr.label])
  File "/usr/local/lib/python2.7/dist-packages/casadi/casadi.py", line 13690, in __init__
    this = _casadi.new_Function(*args)
NotImplementedError: Wrong number or type of arguments for overloaded function 'new_Function'.
  Possible prototypes are:
    Function()
    Function(str)
    Function(Function)
    Function(str,[SX],[SX],dict)
    Function(str,[MX],[MX],dict)
    Function(str,dict:SX,[str],[str],dict)
    Function(str,dict:MX,[str],[str],dict)
    Function(str,[SX],[SX],[str],[str],dict)
    Function(str,[MX],[MX],[str],[str],dict)
  You have: '(str,[MX|str],[MX],[str],[str])'

Some error when trying to run tracking_input_marker

A minor speedup can be achieved in the pseudoinversecontroller if we use SX instead of MX.
Aka, make everything SX functions where possible there.

After implementing the MPC, it seems like lifting the initial conditions rather than taking them in as a parameter is a better option and can cause a speed-up.