Incorporating Unique Control Vectors for Each TaggedWeightedGaussianState in GM-PHD Filter

[AliDehghani88]

Hello,

I am currently working through the Gaussian mixture PHD tutorial and have encountered a challenge that I would like your assistance with.

In the tracker loop, we have current_state = reduced_states, which contains multiple instances of TaggedWeightedGaussianState, each with its state vector, covariance, weight, tag, and timestamp.

During the hypothesis generation step:

hypotheses = hypothesiser.hypothesise(
    current_state,
    measurements,
    timestamp=time,
    order_by_detection=True
)

the Kalman prediction typically uses the formula F*X + B*u. In my case, it's essential to apply distinct control inputs (u) for each TaggedWeightedGaussianState.

X1_k = F * X1_k-1 + B * u1
X2_k = F * X2_k-1 + B * u2

I understand that I can set a uniform control vector as follows example:

hypothesiser.hypothesiser.predictor.control_model.control_vector = np.array([[1], [1.1]])

However, I'm struggling with a scenario where each TaggedWeightedGaussianState in current_state requires a unique control vector.

Could you please guide me on how to assign them?

Thank you

[jmbarr]

I've been having similar thoughts on the shortcomings of the ControlModels which, to be honest, weren't put together with a particular use in mind. I think it's better if you use u as an input to control_model.function() to return the control output. That currently works for LinearControlModel at least but the trouble is nearly all of the predictors will have to be reengineered to use .function(u) instead of .control_vector.