Gazebo City World & Aerostack2 Setup

This week I worked on getting the simulation environment ready — built a custom city world, got drones flying in it, and wired up Aerostack2 as the flight stack.

---

Building My City

The Cat-Mouse exercise needs a visually rich environment, not just a flat ground plane. So I modeled a full urban city in Blender and exported it as a .glb mesh for Gazebo Harmonic. I’m calling it my_city_v3.

I added buildings of varying heights, a road network with highways and intersections, parking lots, a river cutting through the city, trees scattered around, schools, and even cars on the roads. Basically I wanted a miniature city that gives the drones something meaningful to fly around in.

I loaded the mesh as a static model in the SDF world — static so Gazebo doesn’t waste physics cycles on it, but with a collision mesh so drones bounce off buildings if they crash.

---

Drones Spawned

I spawned two quadrotors in the city — one for the cat, one for the mouse. Both use the quadrotor_base model which I set up with four rotors, an IMU with realistic noise parameters, and a forward-facing camera (1280×720, 30 Hz, 90° FOV).

I made the drones namespace-isolated — all cat topics go through /drone_cat/*, all mouse topics through /drone_mouse/*. Only /clock is shared so both drones see the same simulation time.

---

Aerostack2 — Why I Went With It

With drones in the world, the next thing I needed was a flight stack. Raw velocity commands work for quick testing, but for a proper exercise with takeoff sequences, state estimation, and motion control, I needed something more structured.

I went with Aerostack2 — it’s a ROS 2 framework for autonomous multi-aerial-robot systems, developed at Universidad Politécnica de Madrid. It’s natively built on ROS 2, designed for multi-drone scenarios, and has clean modularity — I can swap any component without touching the rest.

Why I picked it for the Cat-Mouse exercise:

• Multi-drone native — built for N drones from the start, not retrofitted
• Platform independent — same code works in sim and on real hardware
• Fully modular — state estimator, motion controller, and platform layer are all swappable plugins
• Clean Python API — students interact through simple calls like takeoff(), get_position(), set_cmd_vel() without touching raw ROS 2 topics

I gave each drone its own independent AS2 stack: a platform layer that bridges to Gazebo, a PID motion controller that tracks velocity references, a state estimator that publishes localization from ground truth, and a behaviors layer that handles arming and takeoff/land sequences. Both stacks run in parallel, fully isolated by namespace.

---

Initial Setup Demo

Here’s a quick video of the initial Gazebo city world with the drones spawned and ready:

---

What’s Next

The simulation substrate is ready — city world loaded, drones spawned, Aerostack2 configured. Next I’ll be working on the chase algorithms and logic — the mouse’s autonomous path generation and the cat’s pursuit controller.