I gain is rounded off.

[nyxrobotics]

Abstract

The problem occurs when adjusting the PID gain of the position.
When the I-gain is 0.4 or less, it is not integrated at all.
When the I-gain is set to 0.5 or higher, it suddenly becomes integrated.

How to Reproduce the Bug

Problem due to the Bug

Factors

Suggestion

[nyxrobotics]

If the value obtained by multiplying the gain by the deviation is less than 1, a digit drop occurs, which occurs for all PID gains.

[nyxrobotics]

The I control calculation was correctly performed with a double type.
However, there seems to be a bug in the calculation of antiwindup.
We need to check if the #72 change has improved the situation.

vesc/vesc_hw_interface/src/vesc_servo_controller.cpp

Lines 103 to 124 in 3d5e62e

	// calculates I control if PD input is not saturated
	if (isSaturated(u_pd))
	{
	u = saturate(u_pd);
	}
	else
	{
	double error_integ_new = error_integ_ + (error_current + error_previous_) / 2.0 * control_period_;
	const double u_pid = u_pd + Ki_ * error_integ_new;
	// not use I control if PID input is saturated
	// since error integration causes bugs
	if (isSaturated(u_pid))
	{
	u = saturate(u_pid);
	}
	else
	{
	u = u_pid;
	error_integ_ = error_integ_new;
	}
	}

[nyxrobotics]

In order to improve the calculation of antiwindup, we referred to the following article.
Anti-windup of PID controller

The behavior was unstable because of the proprietary calculation method. It has been re-implemented using the "Back-Calculation" method.