Disaster Simulation with Drone Navigation (CoppeliaSim Project)

Overview

This project simulates a disaster-struck area filled with fallen trees, rocks, and other obstacles, using CoppeliaSim. A quadcopter (drone) equipped with an RGB-D camera navigates the area. The system supports manual control, dataset collection for AI training, and dynamic environment generation.

Video showcasing the project: Video

---

Main Features

• 🛩️ Drone control (WASD + QE keyboard controls for movement and rotation)
• 📷 RGB-D Camera setup (floating view for live RGB images)
• 🌳 Procedural disaster area generation (rocks, standing/fallen trees)
• 🛠️ Configuration menu to adjust scene parameters at runtime
• 🧠 Event-driven depth dataset collection Depth images for machine learning, triggered by simulation events
• 🎮 Interactive control menus for creating, clearing, restarting scenes
• 📊 Status tab with victim detection visualization including direction indicator, elevation, distance, and signal strength

---

Setup Instructions

1. Install CoppeliaSim 4.9.0.
2. Make sure the zmqRemoteApi plugin is available (standard in 4.6+).
3. Install required Python packages:

pip install coppeliasim-zmqremoteapi-client numpy

4. Start CoppeliaSim with a quadrotor scene containing:

   • Quadcopter model
   • /target dummy

5. Run the simulation from main.py:

python main.py

6. Use the keyboard to interact:
   • Use W/A/S/D + Q/E keys to move and rotate the drone.
   • Press Enter to open the menu.

---

Controls

Key	Action
W	Move Forward
S	Move Backward
A	Strafe Left
D	Strafe Right
Q	Rotate Left (Yaw)
E	Rotate Right (Yaw)
Space	Move Up
Z	Move Down

---

Dataset Collection Behavior

• Event-driven capture on each published simulation/frame event.
• Configurable sampling rate via dataset_capture_frequency (frames per capture).
• Real Euclidean distances to victim (replacing dummy values).
• Immediate one-off capture on victim/detected when distance < victim_detection_threshold.
• Published events:
  • dataset/capture/complete: per-frame metadata { frame, distance, action, victim_vec }
  • dataset/batch/saved / dataset/batch/error: batch save notifications { folder, counter }
  • victim/detected: anomaly alert { frame, distance }
• Batching on in-memory buffers of size batch_size, saved on background thread.
• All event publishes from background threads are thread-safe.
• On shutdown, unsubscribes from dataset events to clean up callbacks.

Data is saved as compressed .npz files in data/depth_dataset/{train,val,test}/batch_XXXXXX.npz containing:

• depths: float32 array (N, H, W)
• poses: float32 array (N, 6)
• frames: int32 array (N,)
• distances: float32 array (N,)
• actions: int32 array (N,)
• victim_dirs: float32 array (N, 4) # (ux,uy,uz,distance)

---

Victim Detection Visualization

The Status Tab provides comprehensive visualization for victim detection:

• Direction Indicator: Radar-like display showing victim's position accurately relative to drone's heading
• Elevation Indicator: Displays victim's height difference in meters with color coding
• Distance Display: Shows distance to victim with color coding (green=near, orange=medium, red=far)
• Signal Strength: Visual indicator that increases as drone gets closer to victim
• Safety Check: System ensures victims spawn at least 2m away from the drone's starting position

All visualizations update in real-time through the event subscription system, with coordinate transformations aligned to the drone's orientation.

---

Future Improvements

• Separate AI module for intelligent navigation (will be hosted in a dedicated repo).
  Placeholder: AI Navigation Repository

• Advanced obstacle avoidance and path planning.

• Dynamic obstacles such as moving objects or agents.

• Expanded disaster scenarios with additional elements.

• General bug fixes and polish.

---

Contribution

• Thomas Lundqvist – Senior Developer
  Main contributor responsible for core development and system architecture.

• Jakub Espandr – Developer & Tester
  Assisted with development and conducted simulation testing.

• Hieu Tran – Documentation & Product Manager
  Handled planning, documentation, and team coordination.

---

This project was built through close collaboration, shared responsibility, and open communication among all team members.

---

License

This project is licensed under the “Personal Use Only License.” See LICENSE for details.