Visibility-Aware Switching Between Radar & LiDAR

Autonomous robots often rely on LiDAR for precise localization and mapping (SLAM), but its performance degrades drastically in low-visibility conditions such as smoke or fog. While radar sensors are robust in such environments, they lack the spatial resolution needed for detailed mapping. To address this trade-off, we propose an adaptive sensor switching framework that dynamically toggles between LiDAR and radar based on environmental visibility. Our method enhances robustness, mapping fidelity, and energy efficiency by using only the most suitable sensor at a given time. Preliminary simulations in CARLA demonstrate the system’s potential for agile, all-weather operation—making it well-suited for disaster response and challenging real-world scenarios.

Visualization

LiDAR Prediction Point Cloud

Radar-LiDAR Point Cloud Fusion

Setup

1. Clone the repository:

   git clone https://github.com/AJaxRUN/Visibility_Aware_Switching_LiDAR_RADAR
   cd Visibility_Aware_Switching_LiDAR_RADAR

2. Install dependencies:

   pip install -r requirements.txt

3. Ensure you have the CARLA simulator installed and running.

Usage

1. Collect Radar and LiDAR Dataset

Run the script to collect radar and LiDAR data:

python collect_radar_lidar_dataset.py

2. Train PointNet Model

Train the PointNet model for point cloud completion:

python train_pointnet.py

3. Visualize Results

Visualize the results using the offline map:

python visualize_offline_map.py

4. Main Entry Point

Run the main script to execute the entire pipeline:

python main.py

Dataset

The dataset/ directory contains .npy files representing LiDAR data collected during simulation. Each file corresponds to a specific frame.

Radar-LiDAR Point Cloud Fusion

Outputs

The output/ directory stores results such as completed point clouds and visualizations generated by the scripts.

Pre-trained Model

The repository includes a pre-trained PointNet model (pointnet_completion.pth) for point cloud completion. You can use this model directly or train your own using the provided script.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgments

• CARLA Simulator for providing a high-fidelity simulation environment.
• PointNet for inspiring the point cloud completion model.