multirotor.optimize

Module Contents

Functions

get_study(→ optuna.Study)	Create an optuna.Study set to maximize reward.
run_sim(→ multirotor.helpers.DataLog)	Run a single episode, where the environment is controlled by the Controller
get_controller(→ multirotor.controller.Controller)	Create a multirotor.controller.Controller object with default values.
make_controller_from_trial(→ dict)	Makes a dictionary of parameters which can be used with the Controller.set_params()
make_env(→ multirotor.env.DynamicsMultirotorEnv)	Make the environment instance to be used by objective(), and set default
make_objective(→ Callable)	Make the function that optuna will use to optimize.
optimize(→ optuna.Study)	Search over parameter space to optimize controller parameters in make_controller_from_trial().
apply_params(→ Dict[str, Dict[str, numpy.ndarray]])	Apply parameters from an optuna.Study to a Controller. Converts the dictionary

Attributes

DEFAULTS
parser

multirotor.optimize.DEFAULTS

multirotor.optimize.get_study(study_name: str = None, seed: int = 0) → optuna.Study

Create an optuna.Study set to maximize reward.

Parameters

study_namestr, optional

If name is given, a study is created in ./studies/study_name.db, by default None

seedint, optional

The random seed to use for parameter sampling, by default 0

Returns

optuna.Study

The study object. Best params can be accessed via study.best_params.

multirotor.optimize.run_sim(env, traj: multirotor.trajectories.Trajectory, ctrl: multirotor.controller.Controller) → multirotor.helpers.DataLog

Run a single episode, where the environment is controlled by the Controller object. The env object must be reset() before calling.

Parameters

envDynamicsMultirotorEnv

The environment to run

trajTrajectory

The waypoints to follow.

ctrlController

The controller which outputs dynamics to be fed to the environment.

Returns

DataLog

A log containing state measurements for each step.

multirotor.optimize.get_controller(m: multirotor.simulation.Multirotor, scurve=False, args: argparse.Namespace = DEFAULTS) → multirotor.controller.Controller

Create a multirotor.controller.Controller object with default values.

Note: max_acceleration, max_velocity, max_error_i etc are important parameters that are set as defaults.

Parameters

mMultirotor

The vehicle to control.

scurvebool, optional

Whether to use an SCurve trajectory planning approach, by default False

argsNamespace, optional

The default settings, by default DEFAULTS

Returns

Controller

multirotor.optimize.make_controller_from_trial(trial: optuna.Trial, args: argparse.Namespace = DEFAULTS, prefix='') → dict

Makes a dictionary of parameters which can be used with the Controller.set_params()

Parameters

trialoptuna.Trial

The trail with the suggest parameters.

argsNamespace, optional

Namespace of optimization arguments, by default DEFAULTS

prefixstr, optional

_description_, by default ‘’

Returns

Dict

A dictionary of parameters to be used by Controller.set_params(**dict)

multirotor.optimize.make_env(vp: multirotor.vehicle.VehicleParams, sp: multirotor.vehicle.SimulationParams, args: argparse.Namespace = DEFAULTS) → multirotor.env.DynamicsMultirotorEnv

Make the environment instance to be used by objective(), and set default env parameters.

Parameters

vp : VehicleParams sp : SimulationParams args : Namespace, optional

Optimization params, by default DEFAULTS

Returns

DynamicsMultirotorEnv

multirotor.optimize.make_objective(vp: multirotor.vehicle.VehicleParams, sp: multirotor.vehicle.SimulationParams, ctrl: multirotor.controller.Controller = None, args: argparse.Namespace = DEFAULTS) → Callable

Make the function that optuna will use to optimize.

Parameters

vp : VehicleParams sp : SimulationParams ctrl: Controller, optional

The controller class to use. If not provided, get_controller is used with DEFAULTS to make a controller instance.

argsNamespace, optional

Optimization params, by default DEFAULTS

multirotor.optimize.optimize(vp: multirotor.vehicle.VehicleParams, sp: multirotor.vehicle.SimulationParams, ctrl: multirotor.controller.Controller = None, ntrials: int = 1000, args: argparse.Namespace = DEFAULTS, seed: int = 0, study_name: str = None, n_jobs: int = 1, timeout: float = None, verbosity=optuna.logging.WARNING) → optuna.Study

Search over parameter space to optimize controller parameters in make_controller_from_trial().

Parameters

vp : VehicleParams sp : SimulationParams ctrl: Controller, optional

The controller class to use. If not provided, get_controller is used with DEFAULTS to make a controller instance.

ntrialsint, optional

Number of trials, by default 1000

argsNamespace, optional

Optimization params, by default DEFAULTS

seedint, optional

Optimization seed, by default 0

study_namestr, optional

Name of study to be saved in studies/study_name.db, by default None

n_jobs: int, optional

Number of parallel optimization jobs to run. If -1, sets to number of processors. By default 1

timeout: float, optional

Number of seconds to run the study for. By default, run until n_trials reached

verbosity_type_, optional

Logging level of optuna.optimize(), by default optuna.logging.WARNING

Returns

optuna.Study

The completed Study object with study.best_params which can be used by apply_params(Controller, **best_params) to get the optimized controller.

multirotor.optimize.apply_params(ctrl: multirotor.controller.Controller, params: dict) → Dict[str, Dict[str, numpy.ndarray]]

Apply parameters from an optuna.Study to a Controller. Converts the dictionary of best_params in the format accepted by controller.set_params()

Parameters

ctrlController

The controller to which to apply the parameters. If None, just return the dictionary of parameters which Controller.set_params() can use.

paramsdict[str, np.ndarray]

The dictionary of parameters returned by optuna.Study

Returns

Dict[str, Dict[str, np.ndarray]]

A nested dictionary for Controller.set_params()

multirotor.optimize.parser