Hydrakon

Getting Started

2 min read

Building the Workspace

Use colcon to build the ROS 2 workspace. We use --symlink-install to allow Python script changes to take effect without rebuilding (mostly).

# Ensure you are in the project root
colcon build --symlink-install

Running the Simulation

To run the simulation, you must set up the environment variables correctly. The PYTHONPATH export is critical for ROS 2 to find the carla module installed in your virtual environment.

# 1. Activate Virtual Environment
source venv/bin/activate
 
# 2. Source the Workspace
source install/setup.bash
 
# 3. Export PYTHONPATH (Crucial for finding 'carla')
export PYTHONPATH=$PYTHONPATH:$(pwd)/venv/lib/python3.12/site-packages
 
# 4. Launch (with optional parameters)
ros2 launch hydrakon_launch launch.py [model_path:=...] [benchmark:=True/False] [manual_control:=True/False]

Launch Arguments

  • model_path: Path to the YOLO model (default: src/.../models/best.onnx).
  • benchmark: Enable inference timing logging (default: False).
  • manual_control: Enable Pygame window for driving the vehicle (default: False).

Examples

Run with Manual Control:

ros2 launch hydrakon_launch launch.py manual_control:=True

Use WASD to drive, Space for handbrake, Q to toggle reverse.

Run with PyTorch model (default) and no benchmarking:

source venv/bin/activate && source install/setup.bash && export PYTHONPATH=$PYTHONPATH:$(pwd)/venv/lib/python3.12/site-packages && ros2 launch hydrakon_launch launch.py

Run with ONNX model and enable benchmarking:

source venv/bin/activate && source install/setup.bash && export PYTHONPATH=$PYTHONPATH:$(pwd)/venv/lib/python3.12/site-packages && ros2 launch hydrakon_launch launch.py model_path:=/home/aditya/HydrakonSimV2/src/hydrakon_camera/hydrakon_camera/best.onnx benchmark:=True

Visualization in RViz

The system publishes 3D markers for detected cones, allowing you to visualize their positions in 3D space.

  1. Start the Simulation: Run one of the launch commands above.
  2. Open RViz2: 
    ros2 run rviz2 rviz2
  3. Configure RViz:
    • Fixed Frame: Set to cone_frame (Type this in manually).
    • Add Display: Click Add → Select MarkerArray → Set Topic to /camera/cone_markers.
    • Add Display (Optional): Click Add → Select Image → Set Topic to /camera/cone_detections_image to see the 2D detections.

The cones will appear as cylinders on the grid (XY plane), color-coded by class:

  • Yellow: Yellow Cone
  • Blue: Blue Cone
  • Orange: Orange Cone / Large Orange Cone
  • White: Unknown

Graph View

Backlinks