core classes

System

class autompc.System(observations, controls, dt=None)

The System object defines a robot system, including the size of the control and observation dimensions and the labels for each control and observation variable.

__init__(observations, controls, dt=None)

Parameters:

• observations (List of strings) – Name of each observation dimension

• controls (List of strings) – Name of each control dimension

• dt (float) – Optional: Data time step for the system.

property controls

Names of each control dimension

property observations

Names of each observation dimension

property ctrl_dim

Size of control dimensions

property obs_dim

Size of observation dimensions

Trajectories

Trajectory

class autompc.Trajectory(system, size, obs, ctrls)

The Trajectory object represents a discrete-time state and control trajectory.

__init__(system, size, obs, ctrls)

Parameters:

• system (System) – The corresponding robot system

• size (int) – Number of time steps in the trajectrory

• obs (numpy array of shape (size, system.obs_dim)) – Observations at all timesteps

• ctrls (numpy array of shape (size, system.ctrl_dim)) – Controls at all timesteps.

property system

Get trajectory System object.

property size

Number of time steps in trajectory

property obs

Get trajectory observations as a numpy array of shape (size, self.system.obs_dim)

property ctrls

Get trajectory controls as a numpy array of shape (size, self.system.ctrl_dim)

TimeStep

class autompc.trajectory.TimeStep(obs, ctrl)

TimeStep represents a particular time step of a trajectory and is returned by indexing traj[i].

obs

Observation. Numpy array of size system.obs_dim

ctrl

Control. Numpy array of size system.ctrl_dim

zeros

autompc.zeros(system, size)

Create an all zeros trajectory.

Parameters:

• system (System) – System for trajectory

• size (int) – Size of trajectory

empty

autompc.empty(system, size)

Create a trajectory with uninitialized states and controls. If not initialized, states/controls will be non-deterministic.

Parameters:

• system (System) – System for trajectory

• size (int) – Size of trajectory

extend

autompc.extend(traj, obs, ctrls)

Create a new trajectory which extends an existing trajectory by one or more timestep.

Parameters:

• traj (Trajectory) – Trajectory to extend

• obs (numpy array of shape (N, system.obs_dim)) – New observations

• ctrls (numpy array of shape (N, system.ctrl_dim)) – New controls

Pipeline

class autompc.Pipeline(system, *components)

The Pipeline class represents a collection of MPC components, including the model, controller, and cost function. A Pipeline can provide the joint configuration space over its constituent components, and can instantiate an MPC given a configuration.

__init__(system, *components)

Parameters:

• system (System) – Corresponding robot system

• components (List of models, controllers, costs and corresponding factories.) – The set of components which make up the pipeline: the model, the controller, and the cost. For each of these components, you can either pass the factory or the instantiated version. For example, you must pass either a Controller or a ControllerFactory, but not both. If the factory is passed, than its hyperparameters will become part of the joint configuration space. If the instantiated version is passed, the component will be treated as fixed in the pipeline.

get_configuration_space()

Return the pipeline configuration space.

__call__(cfg, task, trajs, model=None)

Instantiate the MPC.

Parameters:

• cfg (Configuration) – Configuration from the joint pipeline ConfigurationSpace

• task (Task) – Task which the MPC will solve

• trajs (List of Trajectory) – System ID training set

• model (Model) – A pre-trained model can be passed here which overrides the configuration. Default is None.

Returns:

• controller (Controller) – The MPC controller

• task (Task) – The task with the instantiated cost

• model (Model) – The instantiated and trained model