benchmark-systems

Open collection of benchmark problems. The goal is to provide a common set of models to test and compare different control algorithms. These models have been used by the control community in many different implementations.

The package is designed to be easy to use, using as few dependencies as possible. It is developed by SUPPRESS Research Group, from Universidad de León.

Dependencies

The package has the following dependencies:

• scipy >= 1.14.1
• scipy-dae >= 0.0.6

How it works

Some models are implemented as functions that return a state-space representation of the system:

```python def pendulum(t, x, *, m: float, L: float, drag: float = 0.0, u: float = 0.0) -> np.ndarray: ''' Just a humble pendulum. '''

g = Const.GRAVITY

# State space
dx = np.zeros(2)
dx[0] = x[1]
dx[1] = -g/L * sin(x[0]) - drag/(m*L**2)*x[1] + 1/(m*L**2)*u

return dx

`` These models can be integrated usingsolve_ivpfromscipy`:

```python pendulumstates = solveivp(lambda t, x: pendulum(t, x, m=1, L=2, drag=0.0, u=0), tspan=(0, 20), y0=[np.pi-0.1, 0], teval=np.linspace(0, 20, 1000))

for i, label, in enumerate(['Theta (rad)', 'Omega (rad/s)']): plt.plot(pendulumstates.t, pendulumstates.y[i], label=label) plt.xlabel('Time (s)') plt.legend() ```

Advanced models

Other models can be expressed more conveniently as a set of differential-algebraic equations (DAE). We select scipy-dae to solve these models because it uses almost the same interface as scipy, and aims to be integrated into scipy in the future.

For example, the model of a quadrotor can be expressed as a DAE or an ODE:

Image source: https://es.mathworks.com/help/symbolic/derive-quadrotor-dynamics-for-nonlinearMPC.html

• Newton-Euler model (ODE) States: $x = [x \ y \ z \ \phi \ \theta \ \psi \ s \ v \ w \ p \ q \ r]^T$;

• Euler-Lagrange model (DAE) States: $x = [x \ y \ z \ \phi \ \theta \ \psi \ \dot{x} \ \dot{y} \ \dot{z} \ \dot{\phi} \ \dot{\theta} \ \dot{\psi} ]^T$;

Control actions: $u = [\omega1^2 \ \omega2^2 \ \omega3^2 \ \omega4^2 ]^T$

```python quadrotor_params = { 'Ixx': 1.2, 'Iyy': 1.2, 'Izz': 2.3, 'k': 1, 'L': 0.25, 'm': 2, 'drag': 0.2 }

w2 = quadrotorparams['m']*9.81/(quadrotorparams['k']4) # Squared speed of the propellers in order to hover ... 4kw^2 = mg

u = [w20.97, w2, w21.03, w2] # Pitch only

quadrotorodestates = solveivp(lambda t, x: quadrotorode(t, x, *quadrotorparams, u=u), tspan=(0, 5), y0=[0,0,5] + 9[0], teval=np.linspace(0, 5, 1000)) quadrotordaestates = solvedae(lambda t, x, xdot: quadrotordae(t, x, xdot, **quadrotorparams, u=u), tspan=(0, 5), y0=[0,0,5] + 9*[0], ydot0=np.zeros(12), t_eval=np.linspace(0, 5, 1000))

fig, axs = plt.subplots(2, 1, sharex=True) for i, label in enumerate(['X (m)', 'Y (m)', 'Z (m)', 'Phi (rad)', 'Theta (rad)', 'Psi (rad)']): axs[0].plot(quadrotorodestates.t, quadrotorodestates.y[i], label=label) axs[1].plot(quadrotordaestates.t, quadrotordaestates.y[i], label=label) for i, model in enumerate(['Newton-Euler model', 'Euler-Lagrange model']): axs[i].legend() axs[i].settitle(model) axs[-1].setxlabel('Time (s)') ```

Models

All model simulations are available in the examples folder as a Jupyter notebook.

Installation

Currently, a prerelease version is available via TestPypi:

bash pip install -i https://test.pypi.org/simple/ benchmark-systems