Integrated Fine-Grained 3D Reconstruction and Narrow-Space Planning for Autonomous Parking

ABOUT THIS PROJECT

At a glance

Driving on open roads often involves using broad, simplified methods to identify and outline obstacles through 3D bounding boxes. However, this approach falls short in tight spaces like private garages, public parking facilities, and street parking. In these scenarios, a more detailed reconstruction of obstacles - including vehicles, guardrails, and garage walls - is necessary. Such precision, achieved through camera inputs, is critical for navigating these confined areas safely, enabling effective motion planning and in-car visualization without the risk of collisions. In this project, we propose to construct an integrated framework by jointly designing the following two modules: 1) 3D reconstruction based on novel fine-grained representation of the complex surrounding environments to enable accurate 3D monocular mapping and localization as well as efficient collision check of the downstream planner; 2) Planner in tightly constrained environments utilizing the fine-grained representation constructed from the perception module with efficient computation ensuring collision avoidance and feasibility.