Software-Defined Parallel LiDARs

Active 3D perception with collaborative sensing and dynamic LiDAR scanning.

This project explores collaborative perception and active LiDAR scanning for autonomous driving. The system detects regions of interest from fused multi-vehicle perception and optimizes LiDAR scan parameters to densify foreground point clouds in important areas.

My work included multimodal dataset construction, PointPillars training, pruning and quantization, NVIDIA Orin deployment, TensorRT acceleration, V2V data fusion, global coordinate construction, and dynamic RoI generation.

Highlights:

  • Software-defined LiDAR scanning for active 3D perception.
  • V2V collaborative perception.
  • 10 ms-level TensorRT inference on NVIDIA Orin.

References