As a research intern in HERE North America LLC, I and the team presented a novel collaborative mapping and autonomous parking system for indoor semi-structured multi-level parking garages, based on multi-vehicle perception from 3D lidar. To model the indoor parking space from sensor measurement, a collaborative multi-layered hybrid (CMLH) maps framework is proposed for indoor parking space reconstruction and modeling, and it consists of a map database, collaborative map servers for online global map updating and sharing, and vehicles performing real-time local map perception and modeling.
First an inertial-enhance (IE) generalize iterative closest point (G-ICP) approach is presented to perform registration for lidar odometry (LO), which is coupled with multi-state Extended Kalman Filter (MSEKF) fusion with inertial measurement unit (IMU), and the 3D point cloud model of the parking garage is reconstructed. Then a 2D probabilistic grid map is built from the 3D voxel space and is updated from the real-time lidar perception of the vehicles, which is also used to create a 2D semantic map. Based on our 2D probabilistic grid map and semantic map, a collaborative global path planner is implemented to generate a bidirectional topological graph with A* to explore the best path for navigation. Finally based on the 2D probabilistic grid map, local motion planning is executed using our proposed random exploring algorithm to handle the local navigation with obstacle avoidance and path smoothing. Our pilot experimental evaluation provides a proof of concept for indoor autonomous parking by collaborative perception, map updating and sharing methodology.
The work of this research has been summarized and published in:
- B. Li*, L. Yang*, J. Xiao, R. Valde, M. Wrenn, and J. Leflar. Collaborative Mapping and Autonomous Parking for Multi-story Parking Garage. IEEE Transactions on Intelligent Transportation Systems, Volume: 19, Issue: 5, pp. 1629-1639, 2018.