Tree-SLAM: Localization and Mapping in Dense Forest Environments for Autonomous Vehicles

Automating forestry tasks with intelligent mobile machines requires high-quality localization, which is impossible in dense forests with conventional methods. Accordingly, this paper presents a robust SLAM system tailored for dense forests to provide more accurate and precise localization by exploiting information about the geometric relationship between tree trunk positions. The system consists of three sub-systems: A local tree probability map acts as a short-term memory; a global tree map describes all previously detected trees and the vehicle trajectory using a graphical model; a maps connector that periodically matches geometric triangle features. The developed concept is implemented and evaluated using a simulated model of the Unimog 5023 driving through a forest environment. The system successfully increases localization accuracy and is capable of handling errors.