CarND-Localization-Particle-Filter

Self-Driving Car Engineer Nanodegree Program

---

This repo implements a particle filter localization method for a simulation car with a known map.

Compile and Run

mkdir build && cd build
cmake .. && make
./particle_filter

Implementaion

The implementation follows strictly what has been taught in the lectures.

• void ParticleFilter::init() initializes the particles using a known initial state of the car with randomly-generated noise.
• void ParticleFilter::prediction() uses a simple motion model to predict each particle's state.
• void ParticleFilter::updateWeights() uses a Gaussian cross-correlation between predicted landmark features and observed landmark features to re-weight each particle.
• void ParticleFilter::resample() randmoly re-draws the same number of particles from the exsiting particle pool depending on the weight of each particle; particles with higher weight may be drawn multiple times.
• void ParticleFilter::dataAssociation() uses a nearest-neighbor algorithm to assign observed features to map features. A simple linear search algorithm is used in this implementation. However, since the map is static, a faster search algorithm for this function is to use a K-d tree to represent the map landmarks. I will revisit this in the future.

Result

./run.sh

will give the following result (passed):