Particle Filter Localization for Forklift Simulation

Assignment #4 - Particle Filter Localization

Objectives

The main objective of this project is to localize a forklift in a simulation using a particle filter. The forklift is equipped with a LiDAR and moves in an environment with known landmarks. The particle filter will be used to estimate the position of the forklift throughout the entire simulation.

Implemented Features

• Complete Particle Filter: The filter runs throughout the entire simulation, covering all main steps:

  • Particle initialization
  • State prediction
  • Update based on measurements
  • Particle resampling

• Random Particle Initialization: Particles are initialized randomly within the simulation space.

• Custom Resampling Method: A custom resampling method has been implemented based on the paper "Resampling methods for particle filtering: Classification, implementation and strategies". This improvement aims to optimize the efficiency of resampling and the quality of the estimate.

• Performance Optimization: Several improvements have been made to the code to make the particle filter more efficient in terms of execution and localization accuracy.

Code Structure

The project includes the following main components:

• particle_filter.py: Implements the particle filter with all steps (initialization, prediction, update, resampling).
• main.py: Runs the simulation, executes the particle filter, and saves the results.
• plotter.py: Script to visualize the estimated trajectories against the ground truth data.
• resampling.py: Contains the custom resampling implementation.

Report

The report describes the performance of the particle filter in different scenarios, analyzing the estimated trajectory, accuracy compared to ground truth, and execution times. The output file res.txt contains the X, Y coordinate estimates of the best particle, the corresponding ground truth, and the execution time.

Results are discussed across different scenarios, varying:

• Number of particles
• Sensor errors (LiDAR noise)
• Motion model

Running Instructions

1. Install dependencies:

   pip install -r requirements.txt

2. Run the simulation:

   python main.py

3. Visualize the results: Once the simulation is complete, use the following command to generate the trajectory plots:

   python plotter.py

4. Output:

   • The res.txt file contains position estimates (X, Y) and execution time.
   • The generated plots show the particle filter’s estimated trajectory compared to the forklift’s actual trajectory.

Tested Scenarios

The full report includes analysis of the following scenarios:

1. Scenario 1: Reduced Number of Particles: Basic configuration with a limited number of particles and low sensor noise.
2. Scenario 2: High Sensor Noise: Increased LiDAR noise to test the robustness of the filter.
3. Scenario 3: Increased Number of Particles: More particles used to observe the impact on accuracy and execution times.

Future Improvements

• Further optimization of the resampling algorithm for more complex scenarios.
• Implementation of a more realistic motion model for the forklift.
• Use of parallelization techniques to speed up the particle filter’s execution.

Author

Project developed for the Autonumus Driving course.