Autonomously Following Forest Paths with a Mobile Robot using Semantic Segmentation#

Website: pathy.pfiers.net

This project is Pieter Fiers and Simon Germeau’s professional bachelor’s thesis for UCLL’s Aplied Informatics degree.

We created a rover that is capable of autonomously following forest paths.

We used a Semantic Segmentation CNN, as opposed to the classification CNN used by Giusti et al.1 and Smolyanskiy et al.2. We believe this enables interesting future expansions, like higher-level decision making about path intersections, and the mapping of road geometries.

Repository structure#

The repository is structured according to the four main stages of our project:

Dataprep - Documentation about, and the scripts we used for, the processing and labelling of data used for training.

Model - This folder contains the machine learning process we used to train our CNN.

Rover - Documentation regarding the hardware aspect of our rover.

ROS - All documentation, ROS (Robot Operating System) nodes, and extra files needed to make the rover drive itself.

Presentation#

As pdf or as pptx   source

Demo videos#

All demo videos combined (see YouTube description for timestamps):

youtu.be/SpPcj6MqQEs

Collage of the pathy rover and prediction for a forest path



1

Smolyanskiy et al., “Toward Low-Flying Autonomous MAV Trail Navigation using DeepNeural Networks for Environmental Awareness” May 2017.

2

Giusti et al., “A Machine Learning Approach to Visual Perception of Forest Trails for Mobile Robots” Dec 2015.