Simulation#

Simulation runners and experiment helpers.

Trial Runner#

Quadrotor parameter-estimation trial runner.

Orchestrates a closed-loop simulation: trajectory tracking via LQR, online parameter estimation using any algorithm from online_estimators.estimators, and payload-change events.

online_estimators.simulation.trial.make_estimator(name, theta0, window, seed=None)[source]#

Instantiate an estimator by name.

Parameters:

  • name (str) – Algorithm identifier (case-insensitive). "gt" or "none" returns None.

  • theta0 (np.ndarray, shape (n,)) – Initial parameter estimate.

  • window (int) – Number of stacked measurement rows (used for KF noise sizing).

  • seed (int or None) – If provided, seed the internal RNG of randomised estimators for reproducibility.

Return type:

estimator or None

Raises:

ValueError – If name is not recognised.

online_estimators.simulation.trial.gate_param_update(theta, A, b, rel_resid_max=0.0001, verbose=False)[source]#

Safety gate: accept or reject a proposed parameter update.

Checks physical plausibility (mass bounds, COM bounds, inertia eigenvalues) and numerical quality (relative residual).

Parameters:

  • theta (np.ndarray, shape (10,))

  • A (np.ndarray)

  • b (np.ndarray)

  • rel_resid_max (float)

  • verbose (bool)

Returns:

True if the update should be applied.

Return type:

bool

online_estimators.simulation.trial.run_single_trial(estimator_name, base_seed, seed_offset, est_freq, window, ref_type, T_sim=20.0, noise='none', *, verbose_ctrl=False, verbose_est=False)[source]#

Run one closed-loop quadrotor parameter-estimation trial.

Parameters:

  • estimator_name (str) – Algorithm name (see make_estimator()).

  • base_seed (int) – Base random seed.

  • seed_offset (int) – Offset added to base_seed for reproducibility across trials.

  • est_freq (int) – Estimation update period (in simulation steps).

  • window (int) – Number of consecutive measurement snapshots stacked per update.

  • ref_type (str) – Trajectory type (key into TRAJECTORY_REGISTRY).

  • T_sim (float) – Simulation duration (s).

  • noise (str) – Sensor noise level: "none", "low", "medium", "high".

  • verbose_ctrl (bool) – Print per-step control debug info.

  • verbose_est (bool) – Print per-update estimation debug info.

Returns:

Result dictionary with trajectory data, errors, and metadata.

Return type:

dict

Utilities#

Small numerical helper functions used across simulations.

online_estimators.simulation.utils.abs_residual(A, x, b)[source]#

Absolute residual ||Ax - b||_2.

Parameters:

  • A (np.ndarray, shape (m, n))

  • x (np.ndarray, shape (n,))

  • b (np.ndarray, shape (m,))

Return type:

float

online_estimators.simulation.utils.rel_residual(A, x, b, eps=1e-12)[source]#

Relative residual ||Ax - b||_2 / max(||b||_2, eps).

Parameters:

  • A (np.ndarray, shape (m, n))

  • x (np.ndarray, shape (n,))

  • b (np.ndarray, shape (m,))

  • eps (float) – Floor to avoid division by zero.

Return type:

float

online_estimators.simulation.utils.closest_spd(theta, epsilon=1e-06)[source]#

Project the inertia sub-matrix to the nearest SPD matrix.

Parameters:

  • theta (np.ndarray, shape (10,)) – [m, cx, cy, cz, Ixx, Iyy, Izz, Ixy, Ixz, Iyz]

  • epsilon (float) – Eigenvalue floor.

Returns:

Parameters with inertia projected to SPD.

Return type:

np.ndarray, shape (10,)