Implementation of 2D Motion Planning on an Occupancy Map using RRT Algorithm with Gradient-Based Sampling Guided by Neural Network

Repository Overview

This project implements a Rapidly-exploring Random Tree (RRT) algorithm enhanced with neural network-guided sampling for path planning in 2D environments.

Prerequisites

1. ROS 2 (tested on Humble) with required ros2 packages
2. Python 3.10
3. Required Python packages (requirements.txt):

🛠 Steps to Take

1. Install the required ROS2 packages:

sudo apt-get install ros-humble-nav2-map-server ros-humble-nav2-lifecycle-manager

2. Create ROS 2 workspace:

mkdir -p ros2_ws/src
cd ros2_ws/src

3. Clone the repository:

git clone https://github.com/SirErico/Projekt_MiAPR

4. Before building, create a venv

cd ..
python3 -m venv venv

5. Install the Python dependencies from requirements.txt:

pip install -r requirements.txt

6. Add 'COLCON_IGNORE' file to your venv folder

touch ~/venv/COLCON_IGNORE

7. Source venv

source ~/path/to/venv/bin/activate

8. Build and source workspace

colcon build
source /install/setup.bash

Usage

Training the Neural Network

1. Generate training data:

python3 src/sample_map.py

2. Train the neural network:

python3 src/neural_net.py

Running the RRT Algorithm

To run the RRT algorithm with the neural net model:

ros2 launch mapr_rrt rrt_ai_launch.py model_path:=/your/path/model.keras

---

Scripts:

• map_blur.py: Applies a Gaussian blur to a map image.
• sample_map.py: Samples data points from a map to create a dataset for training a neural network.
• show_map.py: Visualization of the map .pgm and .csv file.
• neural_net.py: Trains a neural network to learn occupancy probabilities from a map dataset.
• show_model.py: Predicts occupancy for the entire map and visualizes the original and predicted occupancy maps.

⚠️ Known Problems

venv Packages Not Visible

If the virtual environment packages are not visible, add the following to your setup.cfg file:

[build]
executable=/usr/bin/env python3