Awesome

Monte Carlo Particle Filter for Localization

Particle Filter Algorithm is a nonparametric implementation of the Bayes Filter to approximate state, for example of a robot moving in a maze.

The idea is to represent the posterior belief by a finite number of random variables (particles). The algorithm is repeatedly resampling those particles based on likelihood derived from a measurement model.

This project is a examination project for SS18 Monte Carlo Methods in Artificial Intelligence and Machine Learning course taught by Prof. Dr. Manfred Opper and Theo Galy-Fajou, at TU Berlin.

[Presentation]

Setup

pip install -r requirements.txt

Usage

> python particle-filter.py

Options
    - scene :  [scene-1, scene-2, scene-1-kidnapping, scene-2-kidnapping, scene-8.12]
    - no_particles : number of particles, for example 100.
    - total_frames :  total time step to run the simulation, it's useful for debugging.
    - show_particles : a boolean option whether to show the particles.
    - no_random_particles: number of random particles introduced to the system, required for kidnapping scenes.
    - save : a boolean option whether to see the simulartor live or save the result as a video.
    - frame_interval : time interval for each frame, default is 50s.

Example

python particle-filter.py --scene scene-1 --no-particles 100 --save

Members

Anders Dahl Hjort, Luis Dreisbach, and Pattarawat Chormai