1. Enhancing Forest Navigation and Mapping
with AI-Controlled Ground Mobile Robots:
Overcoming Challenges of Uneven Terrain
in Digital Forestry
Candidate:
Marco Giberna
Space Robotics Lab
Supervisor:
Dr. Stefano Seriani
Co-Supervisor:
Dr. Matteo Caruso
Academic Year 2022/2023
Computer and Electronic Engineering
Robotics and Artificial Intelligence
3. INTRODUCTION
¡ Monitoring and surveillance of forest environments play a
crucial role in conserving ecosystems and in fighting climate
change
4. INTRODUCTION
¡ Digital Twin: virtual representation of a physical system
¡ Forest Digital Twin: multi-layer representation
¡ Terrain Geometry
¡ Terrain Texture
¡ Terrain Properties
¡ Tree/Plants Segmentation
¡ Tree/Plants Additional Data (e.g. species, height, diameter, vitality, …)
5. INTRODUCTION
Image from Song, J. et al. "A Method for Quantifying Understory Leaf Area Index in a Temperate Forest
through Combining Small Footprint Full-Waveform and Point Cloud LiDAR Data" Remote Sensing, 2021, 13
¡ Forest’s Terrain:
¡ Strongly uneven
¡ Heterogenous
¡ Deformable
6. INTRODUCTION
Data Acquisition Simulation
Environment Creation
Controller Training in the
Simulation Environment
Deployment on
Physical Platform
TERRAIN GEOMETRY +
TEXTURE
TERRAIN DRIVING
CHARACTERISTICS
SCANNING
(IMAGERY
ACQUISITION)
DRIVING AND
MEASURING SIMULATION
ENVIRONMENT
TREE DETECTION +
TRAVERSABILITY ANALYSIS +
PATH PLANNING
7. INTRODUCTION
¡ Traversability: robot’s ability to traverse a given patch of
ground, considering:
¡ Travel Time
¡ Energy Consumption
¡ Risk of Getting Stuck
¡ Risk of Falling
¡ Slippage (difference between perceived and actual position)
¡ …
15. SIMULTANEOUS LOCALIZATION AND MAPPING
¡ Cartographer is an open-source SLAM system
¡ Provides a ROS integration
¡ Supports 2D and 3D map generation
¡ Supports multiple sensors
¡ Requires fine tuning
27. DISCUSSION
¡ Poor results in challenging scenario because of:
¡ Low structurality of the environment
¡ Limited Field of View and vision sensors only looking forward
¡ Low quality camera’s acquistions
¡ Weight imbalance and large vibrations affecting IMUs
28. DISCUSSION
¡ Possible Solutions:
¡ Additional 3D LIDAR or tilted 2D LIDAR facing backwards
¡ Fusing the available IMUs to improve data quality and to get
another odometry source
30. CONCLUSION
¡ Development of 3D SLAM system for the Archimede rover
¡ Implementation of an autonomous navigation system
¡ Datasets collection across various settings, enabling 3D
reconstruction for creating the simulation environment
designed for training mobile robot controllers for forest
navigation and generation of their Digital Twins
¡ Analysis of the performance of the SLAM system