AI ROS 2 Path Planning Activity

This repository contains the ROS 2 support package and A* path-planning interface used in the educational activity described in the WRE 2026 paper draft WRE_2026.pdf, Teaching Artificial Intelligence through Physical Robot Path Planning without Prior ROS Knowledge.

The activity is designed for undergraduate Artificial Intelligence students. Students implement a focused A* planner while the ROS 2 navigation stack, mapping, localization, and robot-control details are handled by the instructor infrastructure. This keeps the programming task centered on search and planning, while still exposing the planner to realistic robotics constraints:

• partially observed occupancy-grid maps;
• obstacle clearance through a distance-transform potential field;
• replanning when unknown map regions become known;
• waypoint simplification for smoother physical robot motion.

Repository Layout

.
├── WRE_2026.pdf                  # Paper draft associated with this repository
├── README.md
└── robot_ai/
    ├── package.xml               # ROS 2 package metadata
    ├── setup.py                  # Python package entry points
    └── robot_ai/
        ├── astar.py              # Student-facing A* planner template
        ├── astar_gab.py          # Reference implementation used by instructors
        ├── map.py                # Occupancy-grid and coordinate conversion helper
        ├── navigator.py          # Nav2 NavigateToPose waypoint runner
        ├── amcl.py               # Pose/map helper used by direct control examples
        ├── goto.py               # Low-level point-to-point controller example
        └── examples/             # PGM maps for local planner tests

System Context

The full classroom setup uses a TurtleBot3 Burger with ROS 2 Humble, Cartographer, and Nav2. The robot publishes an occupancy grid on /map, odometry on /odom, and transforms on /tf. The robot_ai package reads those topics, builds the planner input, asks AStarPathfinder for waypoints, converts map cells back into world coordinates, and sends each waypoint to Nav2 as a NavigateToPose goal in the map frame.

Students do not need to edit ROS launch files or implement ROS nodes. Their main task is to complete the methods in robot_ai/robot_ai/astar.py.

Prerequisites

• ROS 2 Humble
• TurtleBot3 packages
• Nav2
• Cartographer for TurtleBot3
• Python 3
• Python packages used by the planner: numpy, opencv-python, scipy, and matplotlib

For TurtleBot3 simulations or physical robot runs, set the model before starting the navigation stack:

export TURTLEBOT3_MODEL=burger

Build

From a ROS 2 workspace that contains this repository:

colcon build --packages-select robot_ai
source install/setup.bash

If the package is cloned directly as a standalone repository, copy or symlink the robot_ai/ directory into the src/ directory of a ROS 2 workspace before building.

Run With Online SLAM

Start Cartographer:

ros2 launch turtlebot3_cartographer cartographer.launch.py use_sim_time:=True

Start Nav2 using the map topic produced by SLAM:

ros2 launch turtlebot3_navigation2 navigation2.launch.py use_sim_time:=True \
  map:='topic://map'

Run the waypoint navigator:

ros2 run robot_ai navigator

Run With a Saved Map

Save a map:

ros2 run nav2_map_server map_saver_cli -f ~/map

Start Nav2 with the saved map:

ros2 launch turtlebot3_navigation2 navigation2.launch.py map:=$HOME/map.yaml use_sim_time:=True

Then run:

ros2 run robot_ai navigator

Standalone Planner Test

The A* planner can also be exercised without ROS by using the example PGM maps:

cd robot_ai/robot_ai/examples
python ../astar.py

The default main() in astar.py loads map5.pgm, builds an AStarPathfinder, and plots the planned path when a valid implementation is present.

Planner Interface

Students implement these methods in AStarPathfinder:

• preprocess_map(): convert normalized occupancy values into a search map;
• create_potential_field(): penalize cells close to obstacles;
• heuristic(): compute the A* heuristic;
• find_path(): run A* over the occupancy grid;
• reconstruct_path(): recover the path from the predecessor map;
• know_path(): keep only the known and currently executable path prefix;
• simplify_path(): reduce the path into smoother waypoints.

The run() method orchestrates that pipeline and returns a simplified list of map-cell coordinates to the ROS 2 navigation wrapper.

Citation

TODO

License

This project is distributed under the Apache License 2.0. See LICENSE.