# Finding the right PD values for a quadcopter simulator

I am working on this short assignment where I have to find the right PD values for a simulated quadcopter. I have to reach 0.9 meters under 1 second with an overshoot of less than 5%. This is my code:

function [ u ] = pd_controller(~, s, s_des, params)
%PD_CONTROLLER  PD controller for the height
%
%   s: 2x1 vector containing the current state [z; v_z]
%   s_des: 2x1 vector containing desired state [z; v_z]
%   params: robot parameters

% u = 0;

error = s_des - s;
z_des = 0.9;

Kp = 0.5;
Kv = 1.0;

u = params.mass*(z_des + Kp*error(1) + Kv*error(2) + params.gravity);

end


So far this is my output: I've tried many different value combinations and so far this is the closest that I could get.

• Why are you adding z_des to the control? That looks wrong. – holmeski Aug 2 '20 at 21:50
• @holmeski Initially it was in the starter code. Now that I removed it, the drone does not lift off at all. – oo92 Aug 2 '20 at 21:52
• If you set the control to m*(g + I), what's the lowest I can be while still driving the vehicle upwards? – holmeski Aug 2 '20 at 22:04
• z_des seems to be a feedforward value. – jdios Mar 26 at 13:52

This problem can be conveniently solved by resorting to a state-feedback controller as the one depicted below. In this case, we have that $$P=\frac{1}{m \cdot s^2}$$, $$x=\left[ z, \dot{z} \right]^T$$, $$K=\left[k_p, k_d\right]$$, $$k_f=\frac{k_p}{m}$$. Then, we need to apply the feed-forward term $$v=mg$$ to counteract the gravity.

We end up with the transfer function of the closed-loop system:

$$\frac{y(s)}{r(s)}=\frac{\frac{k_p}{m}}{s^2+\frac{k_d}{m}s+\frac{k_p}{m}}.$$

This is a well-known and well-studied second-order system given in the form

$$\frac{\omega_n^2}{s^2+2\xi\omega_ns+\omega_n^2},$$

where the parameters $$\omega_n$$ and $$\xi$$ are expressed as functions of the gains $$k_p$$, $$k_d$$ and the mass $$m$$ as $$\omega_n=\sqrt\frac{k_p}{m}$$ and $$\xi=\frac{k_d}{2\sqrt{m \cdot k_p}}$$.

To meet the given requirements, have a look at the information you can peruse for example at https://www.javatpoint.com/control-system-time-response-of-second-order-system, where the overshoot and the settling-time are provided by the following expressions: $$\begin{cases} e^\frac{\pi\xi}{\sqrt{1-\xi^2}}=0.05 \\ \frac{4}{\xi\omega_n}=1 \end{cases}.$$

The gains $$k_p$$ and $$k_d$$ can be thus readily determined from the formulas above through substitution.

For a unitary mass $$m=1$$ kg, it comes out that $$k_d=8$$ Ns/m and $$k_p=k_f=33.596$$ N/m.

Here's the step response: The step response is generally given with respect to a unitary input step; however, the system is linear and thus this behavior is preserved also for an input step of amplitude $$0.9$$ m.

To conclude, your controller's output shall be:

u = params.mass*params.gravity - kp*z - kd*v_z + kf*z_des;


You are controlling a second order system. Set both kp and kv to 4.

If the system does not reach your desired altitude then there are additional problems with the code.

• It doesn't work. – oo92 Aug 3 '20 at 0:15
• It is likely that there is an error in your setup. Those gains should see the natural frequency to 2 with a damping ratio of 1. Are you sure that you're setting the mass correctly? – holmeski Aug 3 '20 at 0:55
• The code you see above is all I got – oo92 Aug 3 '20 at 1:05
• I recommend checking out my comment to your question. Find the smallest u that will drive the quad off the ground. Start with u=mg then u=mg +.1 and so on until you find some value , let's call it k, that will just drive the quad off the ground. Then try the control law u=m*(kp e(1) + kve(2) +g) + k. That should get you close to the desired error. – holmeski Aug 3 '20 at 2:10
• Also, have you checked all the values to make sure they make sense? Do the desired position and velocity make sense? Is the mass real? – holmeski Aug 3 '20 at 2:13