LiDAR based Graph-SLAM for Autonomous operation in Solar Parks

Abstract

Autonomous vegetation control in solar parks is becoming an increasingly important and relevant field as the number of parks increase with the accelerating need to transition to renewable sources. These parks are characterized by long featureless structures, tall, dense vegetation and poor Global Navigation Satellite System (GNSS) reception. Extensive research has been made into autonomous control in vehicle and industry settings while the challenges involved in adapting these to solar parks remain mostly unknown. The purpose of this work is to utilize a Simultaneous Localization and Mapping (SLAM) algorithm to accurately localize a Husqvarna mower equipped with a LiDAR sensor in the solar park while also creating a map of said park. The implementation should be unaffected by tall vegetation, have minimal drift, create maps of high quality and should not depend on GNSS to achieve this. The SLAM of choice is the cartographer implementation which is a graph-SLAM variant. Furthermore a custom pre-filter, which mainly utilizes RANSAC to detect certain features in the LiDAR data, is designed to improve the localization and mapping results. The proposed solution manages to localize in both high and low vegetation but the accuracy is somewhat subpar with noticeable translational and rotational drift present. The pre-filter manages to improve the quality of the maps by successfully removing the majority of the vegetation, thus reducing the clutter in the generated maps, while the drift remains mostly unimproved. Currently the most efficient way to counter and correct this drift is with the use of GNSS, however this is not a perfect solution due to the aforementioned problems with reception and ideally an alternate solution should be developed. With improvements done in these areas this implementation shows great promise in achieving robust and reliable autonomous control in solar parks and environments of similar characteristics.