witness_controller.h File Reference

#include <stdbool.h>
#include <stdint.h>

Go to the source code of this file.

Classes
struct	witness_coefficients
struct	witness_pid_state
struct	witness_command

Functions
bool	witness_coefficients_valid (const witness_coefficients *coefficients)
void	witness_pid_reset (witness_pid_state *state)
bool	witness_pid_step (const witness_coefficients *coefficients, witness_pid_state *state, uint32_t cycle, int32_t reference, int32_t position, witness_command *command)
bool	witness_lqr_step (const witness_coefficients *coefficients, uint32_t cycle, int32_t reference, int32_t position, int32_t velocity, witness_command *command)

Detailed Description

controllers/c/witness_controller.h.

Definition in file witness_controller.h.

Function Documentation

witness_coefficients_valid()

bool witness_coefficients_valid

(

const witness_coefficients *

coefficients

)

Refuse reversed bounds, bounds excluding zero, negative PID gains or decay outside [0,1].

witness_lqr_step()

bool witness_lqr_step	(	const witness_coefficients *	coefficients,
		uint32_t	cycle,
		int32_t	reference,
		int32_t	position,
		int32_t	velocity,
		witness_command *	command )

Compute u=floor((reference_gain*r-position_gain*y-velocity_gain*v)/2^24) followed by configured output saturation. Gains are supplied from a discrete Riccati design; this kernel does not solve or certify the Riccati equation. Invalid coefficients leave the command unchanged.

witness_pid_reset()

void witness_pid_reset

(

witness_pid_state *

state

)

Reset all PID state, including derivative-on-measurement initialization.

witness_pid_step()

bool witness_pid_step	(	const witness_coefficients *	coefficients,
		witness_pid_state *	state,
		uint32_t	cycle,
		int32_t	reference,
		int32_t	position,
		witness_command *	command )

Compute conditional-integration PID; return false without mutation for invalid coefficients. Error and measurement differences retain their signed 33-bit range. Products accumulate before floor division by 2^24. Output saturation with the prior integral freezes integration if the error would worsen it. Otherwise accept the bounded integral proposal before computing output. The first sample has zero derivative; subsequent samples use a filtered derivative on measurement. No setpoint derivative kick is introduced.