RO47005 PDM: Motion Planning for an Indoor Firefighter Drone

Course project for RO47005 Planning and Decision Making at TU Delft (2024–2025). The task is to plan and fly a quadrotor through a multi-storey building to a target floor. The final code builds a small grid of buildings in PyBullet, computes a global path with either A* on a grid or RRT, and follows that path with a convex model predictive controller (MPC, solved with cvxpy) that constrains the drone to a cylinder around the current path segment.

Team: Oscar Devos, Clara Espirito Santo, Leander Le Ba, Nitya Nanvani

What is in the repository

• Submission/: the final code. path_motion_sim.py is a single script that builds the environment, plans, controls and visualizes. assets/ holds the room URDF and the cf2x drone URDF and mesh.
• project_root/: an earlier, modular version of the project (environment setup, RRT and RRT* planners, a PID controller and trajectory follower). Kept for reference; the submission script does not depend on it.
• dev_test_urdf/: development notebooks and URDF experiments (collision checking, RRT and A* prototypes, an MPC notebook).
• report/: LaTeX source of the report.
• PDM_Group_8_Project_preliminary_report.pdf: the preliminary report.
• PDM_Project.pdf: the course assignment brief.
• POA.txt: the original plan of action. README_old.md: earlier notes on running project_root.

Requirements

There is no requirements file. path_motion_sim.py imports pybullet, numpy, cvxpy, scipy and matplotlib, and uses tkinter for the GUI. Install these into a Python 3 environment before running.

Run

1. Clone the repository and go to the submission folder:

   git clone https://github.com/oscardvs/RO47005_PDM.git
   cd RO47005_PDM/Submission

2. Start the planner and the simulation:

   python3 path_motion_sim.py

   On the first run the script creates a graphs/ directory next to it. A* grids are cached there, so later runs with the same parameters load the grid instead of rebuilding it.

   Four windows open:

   • the PyBullet simulation, with the buildings built and the drone hovering
   • a log window with the time taken to generate or load the A* grid
   • a plot of the drone's current thrust values
   • a small GUI listing the available key commands

3. Control the run from the GUI:

   • F: select a random floor in the building as the goal
   • A: plan with A* and start the flight
   • R: plan with RRT and start the flight
   • Z: re-run A* with the previous goal
   • T: re-run RRT with the previous goal
   • Q: quit

   Both planners can take a while. When planning finishes, the simulation shows the drone flying from start to goal and then opens a pop-up with performance metrics.

4. Tune the simulation. The top of path_motion_sim.py defines global variables with default values: drone mass and thrust limits, obstacle buffers, MPC horizon, RRT iterations, A* cell size, number and height of buildings, random seed, and camera options. Edit them to change the run.