MCL

class MCL(Simulator)

The Monte Carlo Localization algorithm implementation. Inherits the Simulator.

This implementation was not done to be general, is specific to a bi-dimensional squared grid map.

The noises are Simulator class attributes and are initialized with it default values. If you want to change the noises values change it using the Simulator and not MCL. Also, be sure to set a new map to trigger the pre-processing steps which uses the noise values. The default values:

  • Simulator.p_hit = 0.9: The probability of the measurement to be right.
  • Simulator.p_miss = 0.1: The probability of the measurement to be wrong.
  • Simulator.p_exact = 0.8: The probability of the robot perform an exact motion.
  • Simulator.p_overshoot = 0.1: The probability of the robot perform more than desired motion.
  • Simulator.p_undershoot = 0.1: The probability of the robot perform less than desired motion.

See also the Simulator docs for more details.

__init__

 | __init__(map_: Map)

Construct the MCL object. It starts an uniform probability distribution and randomly set the robot simulated position. Also, do the pre-processing to build the sense and movement matrices.

Arguments:

  • map_: The map required by the Monte Carlo Localization.

move

 | move(cmd=-1) -> int

Do the movement operation at Monte Carlo Localization. It can receive an external command (cmd) in the form:

  • 0: Move left. Prefer to use Simulator.left_cmd.
  • 1: Move right. Prefer to use Simulator.right_cmd.
  • 2: Move up. Prefer to use Simulator.up_cmd.
  • 3: Move down. Prefer to use Simulator.down_cmd.

Or if the number is negative use a navigation policy:

  • -1: Random movements (default)
  • -2: Minimize the probability difference between places with and without landmarks (see articles).
  • -3: Minimize the expected quadratic sum. Not in the articles but also a good one (see project docs).

Arguments:

  • cmd: The desired movement command.

Returns:

The desired movement command.

sense

 | sense(sens: bool)

Do the Monte Carlo Localization measurement update. Also, it normalize the probability distribution.

Arguments:

  • sens: The sensor reading.

estimated_position

 | @property
 | estimated_position()

Returns:

The Monte Carlo Localization estimated position. None if there aren't one yet.

prob_distribution

 | @prob_distribution.setter
 | prob_distribution(x)

Set the probability distribution with a copy of the new one.

Arguments:

  • x: New probability distribution.

map

 | @map.setter
 | map(x: Map)

Set the map with the new one (isn't a copy). Also, updates the sensor and movement matrices. Finally, set the real position to a random one.

Arguments:

  • x: The new map.

uniform_distribution

 | @property
 | uniform_distribution() -> np.array

Returns:

The uniform probability distribution of the actual map

reset_distribution

 | reset_distribution()

Reset the probability distribution to the uniform one. Automatically called when changing or setting the map.

Should be called after setting the real_position unless you want to analyse the kidnapped robot problem.