Title:Algorithm–Hardware Co-Design for LiDAR Point Cloud Processing: Classic and Learning-Based Approaches Abstract:LiDAR perception is a cornerstone of modern autonomous systems, enabling reliable 3D scene understanding for self-driving, robotics, and intelligent infrastructure. As sensors shift from mechanical to solid-state designs and produce denser, higher-rate point clouds, computational demands escalate. Meeting real-time, low-power, and small-form-factor requirements on edge platforms remains challenging for conventional CPU/GPU pipelines.This dissertation presents a hardware-friendly LiDAR perception framework through algorithm–hardware co-design on FPGAs and introduces three key designs: (1) a stream-based ground-segmentation method with a parallel optimization path for solid-state LiDAR, enabling fully pipelined, low-latency processing; (2) a Fast Channel-Clustering (FCC) algorithm for real-time instance formation using line-buffered, local-search micro-architectures; and (3) a ReAct Binary PointPillar detector that binarizes network weights and replaces multiplications with XNOR–popcount, substantially reducing computational and memory requirement while preserving accuracy. All designs emphasize deep pipelining, resource efficiency, and deterministic execution to meet tight timing and energy budgets on edge platforms.In addition to standalone modules, this work integrates classical geometric methods with learning-based approaches to form a hybrid pipeline. It proposes a panoptic-segmentation flow that couples semantic segmentation with clustering-based instance formation, achieving state-of-the-art accuracy.The proposed designs are evaluated on KITTI/SemanticKITTI dataset and real-world collected LiDAR data, demonstrating strong accuracy, high throughput, and robust generalization across multiple types of LiDAR — pointing toward practical, deployable LiDAR intelligence at the edge. Research Advisor:Prof. Xinming HuangECE Department, WPI Committee Members:Prof. Shahin TajikECE Department, WPIProf. Tian Guo Computer Science Department, WPIProf. Honggang WangGraduate Department Chair, Computer Science & Engineering, Yeshiva University