Unit 4

Probabilistic Map-Based Localization

(PMBL)
Probabilistic Map-Based Localization (PMBL) is a concept in mobile robotics where a robot uses
probabilistic techniques to determine its location within a given map. This approach helps the robot
localize itself in an environment by considering uncertainty in its sensory data and in its movement.
Localization is essential in autonomous navigation since it allows the robot to understand "where it is"
in order to plan "where to go."
Key Concepts in Probabilistic Map-Based Localization
1. Map Representation: The robot needs a map of the environment, which can be a:

• Grid Map: Divides the space into a grid of cells, where each cell indicates whether it's
occupied or free.
• Topological Map: Uses nodes and edges to represent possible positions and pathways.
• Feature-Based Map: Represents only certain landmarks or features in the environment, like
doors, corners, or objects.
2. Probabilistic Representation of Position: In real-world environments, sensors and actuators
are subject to noise and errors, making it impossible for a robot to know its exact position.
Probabilistic methods provide a way to estimate position based on likelihood. Two main
components are used:

• Belief: The robot’s estimation of its location, represented as a probability distribution over all
possible locations.
• State Space: All possible positions and orientations (often called “poses”) of the robot in the
environment.
3. Bayesian Filtering: Bayesian filtering, specifically through algorithms like Markov
localization or Kalman filters (for continuous state spaces), is used to update the robot’s
belief about its location over time. The steps are as follows:

• Prediction Step: Uses the robot's motion model to predict where it might be after moving.
• Update Step: Uses sensory data to update its belief about its position.
4. Sensor and Motion Models:

• Motion Model: Predicts the robot's position after it moves based on control inputs (e.g.,
speed, turn angle). Because movement is imperfect, the motion model includes uncertainty,
typically represented by a probability distribution.
• Sensor Model: Updates the robot's belief about its location based on sensor readings (e.g.,
lidar, sonar, camera). It calculates the likelihood of observing the current sensor data from
each possible location on the map.
5. Monte Carlo Localization (Particle Filter): Monte Carlo Localization (MCL) is a widely
used PMBL technique that represents the robot's belief using a set of weighted particles, each
representing a possible pose. MCL has several steps:
 • Prediction: Moves particles according to the motion model, adding random noise.
• Correction: Updates the weight of each particle based on the likelihood of the sensor data
given the particle’s location.
• Resampling: Particles with higher weights are duplicated, while those with lower weights are
discarded, focusing the estimation on the most probable areas.
6. Global Localization and Kidnapped Robot Problem:

• Global Localization: The process of determining a robot's position without prior knowledge,
which involves evaluating possible locations across the entire map.
• Kidnapped Robot Problem: A challenge where the robot is suddenly relocated to an
unknown place. Probabilistic methods like MCL are designed to handle this by periodically
introducing random particles across the map to ensure recovery if the robot's estimated
position becomes highly uncertain.
Advantages of Probabilistic Map-Based Localization
• Handles Uncertainty: Probabilistic methods can account for noise in sensor and motion data,
making them robust to real-world conditions.
• Adaptability: They are flexible across different environments and can adapt to changes in
sensor readings or map updates.
• Recovery from Errors: Approaches like MCL allow for recovery from localization errors,
such as those introduced by environmental changes or sensor failures.
Applications in Mobile Robotics
PMBL is used in various mobile robot applications, including:
• Autonomous Vehicles: To navigate roads with uncertain and dynamic elements.
• Service Robots: For indoor navigation in environments like hospitals or warehouses.
• Search and Rescue Robots: Where environments are often unfamiliar or partially obstructed.

Landmark-Based Navigation
Landmark-Based Navigation enables mobile robots to navigate by recognizing fixed features or
landmarks in an environment, using these as reference points to localize and navigate effectively. This
method is particularly useful in complex settings where conventional navigation aids like GPS may be
unavailable or unreliable.
Key Components of Landmark-Based Navigation
1. Landmarks
Landmarks are distinctive, stationary features that serve as reference points for localization.
They can be:

• Natural Landmarks: Existing objects such as trees, doors, or furniture.

• Artificial Landmarks: Purposefully placed markers like QR codes or colored tags for easy
detection.
Effective landmarks are unique, stable, and detectable by the robot’s sensors, which allows for
accurate navigation without ambiguity.
2. Detection and Recognition
Robots use sensors to detect and recognize landmarks:

• Computer Vision: Identifies visual cues like shapes, colors, or patterns.

• Lidar and Sonar: Detects structural features like walls or corners.
• RFID/Bluetooth: Detects artificial markers often used in indoor spaces.
3. Localization Using Landmarks
Landmark-based localization allows the robot to estimate its position by measuring its relative
distance and angle to known landmarks:

• Pose Estimation: Calculates the robot’s position and orientation based on the detected
landmarks.
• Map Correlation: Compares observed landmarks with a map of known locations to improve
localization accuracy.
4. Probabilistic Approaches
Due to sensor noise and imperfect movement, probabilistic methods like Extended Kalman
Filters (EKF) and Particle Filters are often used:

• EKF: Fuses data to update the robot’s estimated position based on detected landmarks.
• Particle Filters: Represents possible locations as particles, adjusting them based on observed
landmarks.
5. Navigation Process
The robot’s navigation involves:

• Route Planning: Planning a path that includes key landmarks to reach a destination.
• Continuous Recalibration: Regularly updating its position based on detected landmarks.
• Error Correction: Adjusting its route if discrepancies between detected and expected
landmarks occur.
6. Landmark Association and Ambiguity Resolution
In environments with multiple similar landmarks, the robot uses data association techniques
to match detected landmarks with those in its map. Techniques like Nearest Neighbor
Association or Joint Compatibility Tests help avoid ambiguity and improve accuracy.
Benefits of Landmark-Based Navigation
• Reliable in GPS-Denied Environments: Ideal for indoor or obstructed environments where
GPS is unavailable.
• Accurate Localization: Continuous reference to landmarks reduces positional drift.
• Real-Time Error Correction: Frequent recalibration ensures precise navigation even in
changing or cluttered environments.
Applications
• Warehouse Automation: Using QR codes or RFID tags as landmarks to guide robots through
organized aisles.
• Indoor Service Robots: Reliance on furniture or walls as landmarks for navigating within
buildings.
• Autonomous Vehicles: Using natural or artificial landmarks on urban streets where GPS may
be unreliable.
• Search and Rescue: Navigating in unfamiliar or cluttered spaces by recognizing visible
features.