mirror of https://github.com/ArduPilot/ardupilot
uncrustify libraries/AC_PID/AC_PID.h
This commit is contained in:
parent
66ab46fa88
commit
7aea7dc825
|
@ -7,114 +7,134 @@
|
|||
#define AC_PID_h
|
||||
|
||||
#include <AP_Common.h>
|
||||
#include <math.h> // for fabs()
|
||||
#include <math.h> // for fabs()
|
||||
|
||||
/// @class AC_PID
|
||||
/// @brief Object managing one PID control
|
||||
class AC_PID {
|
||||
public:
|
||||
|
||||
/// Constructor for PID that saves its settings to EEPROM
|
||||
///
|
||||
/// @note PIDs must be named to avoid either multiple parameters with the
|
||||
/// same name, or an overly complex constructor.
|
||||
///
|
||||
/// @param initial_p Initial value for the P term.
|
||||
/// Constructor for PID that saves its settings to EEPROM
|
||||
///
|
||||
/// @note PIDs must be named to avoid either multiple parameters with the
|
||||
/// same name, or an overly complex constructor.
|
||||
///
|
||||
/// @param initial_p Initial value for the P term.
|
||||
/// @param initial_i Initial value for the I term.
|
||||
/// @param initial_d Initial value for the D term.
|
||||
/// @param initial_imax Initial value for the imax term.4
|
||||
///
|
||||
AC_PID(
|
||||
const float &initial_p = 0.0,
|
||||
const float &initial_i = 0.0,
|
||||
const float &initial_d = 0.0,
|
||||
const int16_t &initial_imax = 0.0)
|
||||
{
|
||||
_kp = initial_p;
|
||||
_ki = initial_i;
|
||||
_kd = initial_d;
|
||||
_imax = abs(initial_imax);
|
||||
}
|
||||
///
|
||||
AC_PID(
|
||||
const float & initial_p = 0.0,
|
||||
const float & initial_i = 0.0,
|
||||
const float & initial_d = 0.0,
|
||||
const int16_t & initial_imax = 0.0)
|
||||
{
|
||||
_kp = initial_p;
|
||||
_ki = initial_i;
|
||||
_kd = initial_d;
|
||||
_imax = abs(initial_imax);
|
||||
}
|
||||
|
||||
/// Iterate the PID, return the new control value
|
||||
///
|
||||
/// Positive error produces positive output.
|
||||
///
|
||||
/// @param error The measured error value
|
||||
/// @param dt The time delta in milliseconds (note
|
||||
/// that update interval cannot be more
|
||||
/// than 65.535 seconds due to limited range
|
||||
/// of the data type).
|
||||
/// @param scaler An arbitrary scale factor
|
||||
///
|
||||
/// @returns The updated control output.
|
||||
///
|
||||
int32_t get_pid(int32_t error, float dt);
|
||||
int32_t get_pi(int32_t error, float dt);
|
||||
int32_t get_p(int32_t error);
|
||||
int32_t get_i(int32_t error, float dt);
|
||||
int32_t get_d(int32_t error, float dt);
|
||||
/// Iterate the PID, return the new control value
|
||||
///
|
||||
/// Positive error produces positive output.
|
||||
///
|
||||
/// @param error The measured error value
|
||||
/// @param dt The time delta in milliseconds (note
|
||||
/// that update interval cannot be more
|
||||
/// than 65.535 seconds due to limited range
|
||||
/// of the data type).
|
||||
/// @param scaler An arbitrary scale factor
|
||||
///
|
||||
/// @returns The updated control output.
|
||||
///
|
||||
int32_t get_pid(int32_t error, float dt);
|
||||
int32_t get_pi(int32_t error, float dt);
|
||||
int32_t get_p(int32_t error);
|
||||
int32_t get_i(int32_t error, float dt);
|
||||
int32_t get_d(int32_t error, float dt);
|
||||
|
||||
|
||||
/// Reset the PID integrator
|
||||
///
|
||||
void reset_I();
|
||||
/// Reset the PID integrator
|
||||
///
|
||||
void reset_I();
|
||||
|
||||
/// Load gain properties
|
||||
///
|
||||
void load_gains();
|
||||
/// Load gain properties
|
||||
///
|
||||
void load_gains();
|
||||
|
||||
/// Save gain properties
|
||||
///
|
||||
void save_gains();
|
||||
/// Save gain properties
|
||||
///
|
||||
void save_gains();
|
||||
|
||||
/// @name parameter accessors
|
||||
//@{
|
||||
/// @name parameter accessors
|
||||
//@{
|
||||
|
||||
/// Overload the function call operator to permit relatively easy initialisation
|
||||
void operator() (const float p,
|
||||
const float i,
|
||||
const float d,
|
||||
const int16_t imaxval) {
|
||||
_kp = p; _ki = i; _kd = d; _imax = abs(imaxval);
|
||||
}
|
||||
/// Overload the function call operator to permit relatively easy initialisation
|
||||
void operator () (const float p,
|
||||
const float i,
|
||||
const float d,
|
||||
const int16_t imaxval) {
|
||||
_kp = p; _ki = i; _kd = d; _imax = abs(imaxval);
|
||||
}
|
||||
|
||||
float kP() const { return _kp.get(); }
|
||||
float kI() const { return _ki.get(); }
|
||||
float kD() const { return _kd.get(); }
|
||||
int16_t imax() const { return _imax.get(); }
|
||||
float kP() const {
|
||||
return _kp.get();
|
||||
}
|
||||
float kI() const {
|
||||
return _ki.get();
|
||||
}
|
||||
float kD() const {
|
||||
return _kd.get();
|
||||
}
|
||||
int16_t imax() const {
|
||||
return _imax.get();
|
||||
}
|
||||
|
||||
void kP(const float v) { _kp.set(v); }
|
||||
void kI(const float v) { _ki.set(v); }
|
||||
void kD(const float v) { _kd.set(v); }
|
||||
void imax(const int16_t v) { _imax.set(abs(v)); }
|
||||
void kP(const float v) {
|
||||
_kp.set(v);
|
||||
}
|
||||
void kI(const float v) {
|
||||
_ki.set(v);
|
||||
}
|
||||
void kD(const float v) {
|
||||
_kd.set(v);
|
||||
}
|
||||
void imax(const int16_t v) {
|
||||
_imax.set(abs(v));
|
||||
}
|
||||
|
||||
float get_integrator() const { return _integrator; }
|
||||
void set_integrator(float i) { _integrator = i; }
|
||||
float get_integrator() const {
|
||||
return _integrator;
|
||||
}
|
||||
void set_integrator(float i) {
|
||||
_integrator = i;
|
||||
}
|
||||
|
||||
static const struct AP_Param::GroupInfo var_info[];
|
||||
static const struct AP_Param::GroupInfo var_info[];
|
||||
|
||||
private:
|
||||
AP_Float _kp;
|
||||
AP_Float _ki;
|
||||
AP_Float _kd;
|
||||
AP_Int16 _imax;
|
||||
AP_Float _kp;
|
||||
AP_Float _ki;
|
||||
AP_Float _kd;
|
||||
AP_Int16 _imax;
|
||||
|
||||
float _integrator; ///< integrator value
|
||||
int32_t _last_input; ///< last input for derivative
|
||||
float _last_derivative; ///< last derivative for low-pass filter
|
||||
float _output;
|
||||
float _derivative;
|
||||
float _integrator; ///< integrator value
|
||||
int32_t _last_input; ///< last input for derivative
|
||||
float _last_derivative; ///< last derivative for low-pass filter
|
||||
float _output;
|
||||
float _derivative;
|
||||
|
||||
/// Low pass filter cut frequency for derivative calculation.
|
||||
///
|
||||
static const float _filter = 7.9577e-3; // Set to "1 / ( 2 * PI * f_cut )";
|
||||
// Examples for _filter:
|
||||
// f_cut = 10 Hz -> _filter = 15.9155e-3
|
||||
// f_cut = 15 Hz -> _filter = 10.6103e-3
|
||||
// f_cut = 20 Hz -> _filter = 7.9577e-3
|
||||
// f_cut = 25 Hz -> _filter = 6.3662e-3
|
||||
// f_cut = 30 Hz -> _filter = 5.3052e-3
|
||||
/// Low pass filter cut frequency for derivative calculation.
|
||||
///
|
||||
static const float _filter = 7.9577e-3; // Set to "1 / ( 2 * PI * f_cut )";
|
||||
// Examples for _filter:
|
||||
// f_cut = 10 Hz -> _filter = 15.9155e-3
|
||||
// f_cut = 15 Hz -> _filter = 10.6103e-3
|
||||
// f_cut = 20 Hz -> _filter = 7.9577e-3
|
||||
// f_cut = 25 Hz -> _filter = 6.3662e-3
|
||||
// f_cut = 30 Hz -> _filter = 5.3052e-3
|
||||
};
|
||||
|
||||
#endif
|
||||
|
|
Loading…
Reference in New Issue