/carpet_localisation

Primary LanguageJupyter Notebook

Carpet localisation

For a full project overview, see the wiki!

A particle filter implementing carpet-based localisation.

Inputs to particle filter are:

  • odometry: pose delta since previous update
  • color: current detected carpet color under robot
  • map: grid map of carpet colors

Output of particle filter is the current pose (2D) of the robot

simulation_playback

Figure: playback of particle filter using simulated input data (see simulator.py), driving in a loop on a small map. Legend:

  • green arrow: ground truth pose
  • large red arrow: estimated pose from particle filter
  • small red arrows: particle filter particles
  • blue/grey/beige background: carpet color pattern map

realworld test

Figure: localisation test in target environment (carpeted office).

For ROS wrapper around this package, see carpet_localisation_ros.

For classifier trained to distinguish the four colors of carpet in the target environment, see carpet_color_classification

Note: for an overview of the broader project, see the wiki.

Dev setup:

git clone git@github.com:tim-fan/carpet_localisation.git
python3 -m venv venv
source venv/bin/activate
pip install -r requirements-dev.txt
pip install -e .

Run tests

pytest .

Or to view output of plotting tests:

pytest . --show_plot True