ADD: PID controller class - #88
Conversation
| import numpy as np | ||
|
|
||
| class PID: | ||
| def __init__(self, P_coeff: float = 0.0, I_coeff: float = 0.0, D_coeff: float = 0.0, integration_type: str = "R_rect"): |
There was a problem hiding this comment.
This is nitpicky but for consistency with the usual controls convention I'd like to see these coefficients changed to K_P, K_I, K_D and referred to as gains in the documentation. It'll make the code more searchable by adhering to commonly used nomenclature.
| """ | ||
| return tuple(self.PID_coeffs) | ||
|
|
||
| def begin_sim(self, starting_state_vec: np.ndarray, desired_state_vec: np.ndarray, default_timestep: float = 1.0, init_time: float = 0.0): |
There was a problem hiding this comment.
I don't think I see the need for this. We're going to provide the current state every time we call step() anyway and the setpoint is something we should be able to change at any time. As far as init_time, if we want to use absolute time then you can just specify that in init(). To be honest though it shouldn't really matter because we only need to know the elapsed time between steps. We should be able to just initialize the class with its gains and run the computation at each timestep.
| self.integrated_error = np.zeros(np.shape(self.goal)[0]) | ||
| self.prev_integ_error = self.integrated_error.copy() | ||
|
|
||
| def _integrate_error(self, error: np.ndarray, prev_error: np.ndarray, step_duration: float): |
There was a problem hiding this comment.
I suspect this approach of directly modifying self.integrated_error is going to cause problems when used in an adaptive-step simulation integrator. I think this might need to be handled at a higher level such that the integrated term is only updated after the simulation has committed to a timestep. I don't love having to explicitly update outside the controller's own context, but I don't see another way right now. The result would be that you need a sort of candidate integrated error for the timestep and then a function that locks it in when the next simulation timestep is taken.
| step_duration = self.prev_timestep | ||
| self.integrated_error = self.prev_integ_error.copy() | ||
|
|
||
| self.error = self.goal - state_vec |
There was a problem hiding this comment.
Please add a function to update self.goal. And I would prefer to refer to it as self.setpoint or self.target_state
2 participants
This PR contains code for a PID controller class.