The use of physical vehicles and environments during vehicle research and development is highly resource-intensive, particularly for autonomous vehicles. Recently, digital models are therefore increasingly used instead, which require high levels of fidelity and validity. While the two aforementioned qualities are often lacking, an absence of versatility for multi-purpose use is even more prevalent in current digital models. In response to these challenges, this work presents a novel real-time multi-physics digital twin of an offroad vehicle with high levels of fidelity and validity, both regarding the vehicle dynamics and hydraulics, as well as regarding the visual representation of the environment and the exteroceptive sensor emulation. The versatility of the digital twin enables its usage for vehicle development tasks concerning mechanical components and driveline, as well as for visual machine learning tasks, such as generation of auto-annotated visual training data. Development of control algorithms leveraging both visual input and mechanical systems is also enabled. Furthermore, the real-time capability allows for Hardware-in-the-Loop and Vehicle-in-the-Loop simulation. The modeling, calibration, and real-world validation of the digital twin is presented, with an emphasis on the vehicle dynamics and hydraulics. The shown validity enables advancements in the development of autonomous offroad vehicles.

In Finland and Scandinavia, even-aged forest management predominates, often including mechanical site preparation and manual planting. Growing labor shortages and increased demand for sustainability have driven interest in mechanized and autonomous planting systems. This study evaluates two automated Coverage Path Planners (CPP), Pathfinder and TerraTrail, developed to optimize planting routes for mechanized forest regeneration. Their performance is compared to the routes of the manually operated mechanized planting machine, PlantMax. Three operational sites in Sweden, representing varied terrain and hydrological conditions are evaluated. The evaluation focuses on coverage, Euclidean and Dubins path lengths. Both CPPs incorporate Digital Elevation Models (DEM), Depth-to-Water (DTW) maps and vehicle-specific kinematics to generate planting routes. Two scenarios are evaluated: one where the CPPs neglect the DTW map, and another where the CPPs are constrained to avoid DTW values below 0.3 m. Results show that automated CPPs achieve 15–19% higher coverage than manual planning on average. Pathfinder showed similar normalized path lengths in an unconstrained scenario as the manual operator, but 14% shorter in the constrained environment. TerraTrail shows 7% longer normalized path lengths in an unconstrained scenario, while the constrained scenario shows similar path lengths as the manual operator. These findings emphasize the potential of deploying automated CPP systems to enhance precision, sustainability, and labor efficiency of silvicultural operations. The CPPs support both autonomous deployment and decision support tool for operators. Further refinement, including combining both CPPs to leverage the best functions of each, along with reversible path planning, could enhance their value in forestry practices.

This dataset was collected during the summer of 2023 as part of a forestry field study in southern Sweden. Data was gathered from 34 replanted clear-cut sites within three forest management areas operated by Södra, located near Norrköping, Kinna, and Växjö. A total of 2,780 seedlings were documented, resulting in 14,992 images. Each seedling was photographed from above and from the side in four different orientations.

The majority of seedlings are Norway spruce (Picea abies) and Scots pine (Pinus sylvestris), with a smaller number of silver birch (Betula pendula) and European larch (Larix decidua). Most (14,748) side-view images include a corresponding left-lens stereo image.

Seedling condition, height, and planting quality were assessed in the field by a single technician. Plant height was measured using a ruler, and planting angle was visually estimated with the aid of a protractor. A measurement frame placed in the top-down images provided scale and directional reference to support these assessments.

Site-level metadata, such as soil characteristics was retrieved via API from the Swedish Forest Agency (Skogsstyrelsen). Prior forest treatment was also added based on site preparation documents.

Object detectors for autonomous forestry operations have previously been developed mainly by training on physical manually annotated data, which is both time-consuming and costly. Since the ground truth in the virtual model is known, the training data can be auto-annotated, enabling the creation of larger training datasets, while also improving time and cost efficiency. In this work, a virtual environment in Unity is used in co-simulation with a real-time digital twin of a physical forestry vehicle, to generate realistic auto-annotated training data, as captured by an onboard stereo camera. First, it is shown that a log detector trained on physical data can detect logs in the virtual environment. Second, new detectors are trained, using different shares of virtual and physical data. It is shown that a detector trained using only virtual data, can learn to detect logs in the physical world. Moreover, virtual pre-training is shown to improve the performance of physically trained and tested detectors, both at low availability of physical training data, and in terms of domain generalization. A detailed detector performance analysis also highlights further potential and opportunities for future improvements. Furthermore, the real-time capable virtual models enable future machine learning tasks utilizing different levels of Hardware-in-the-Loop. 

Manual or mechanized tree planting typically follows final felling to ensure satisfactorilyregrowth of the new forest stand. As part of quality assurance, inventories are conductedsome time after planting to verify properly planted seedlings and to gather forest standplanting statistics, which are both costly and time consuming. With the advent of AI andfaster image processing, several methods are now available that can automate theassessment of replanted tree seedlings and objectively quantify planting quality. In thisstudy, we used a dataset consisting of over 3000 stereo images and corresponding depthmaps, annotated with seedling length, position and tilt, as well as classifications of plantingspots, site preparation methods, and many other parameters relevant for qualityassessment. Using computer vision, we evaluated both rule-based approaches incombination with trained AI models to find suitable methods for automating plantingquality assessment. The annotated dataset was used to validate the different estimationmethods. First, we developed a model that automatically detected seedlings in image data.The outcome was used to evaluate the presence of vital, living seedlings. Next, we were ableto discern whether the planting spot were positioned favourably using depth data as input.These findings suggest that image and depth data can be utilized to assess planting qualityin a stand, either after manual planting or during mechanized planting, providing valuablefeedback to the system or operator. Drone overflights may also serve as a suitable methodfor this assessment. In recent years, research has indicated that autonomous forestregeneration may be achievable in the near term. The results of this study may acceleratethis development, as planting quality assessment is essential when such autonomoussystems operate independently of human oversight.

This study presents a metaheuristic approach to coverage path planning for ground-based forest operations, focusing on minimizing path lengths for forest vehicles while considering terrain characteristics and vehicle parameters. Forest vehicles can usually tolerate higher pitch than roll angles, which makes them vulnerable to rollover. To mitigate that, this method utilizes a genetic algorithm to optimize the sequence of nodes, which are scattered over the site with equal spacing. The coverage path planner then calculates the Dubins path distance between every node in the fitness function, together with penalties for exceeding pitch, roll and soil moisture constraints for the vehicle. This ensures that the path planner tries to make the most traversable path as possible, while trying to minimize the driving distance. Two synthetic test sites resembling primitive challenging terrains, and one real site were utilized to theoretically evaluate the proposed method. The results show that aligning the node patterns with the critical slope headings, instead of having a straight pattern, had little effect on the path length. However, square grids can yield shorter paths across multiple runs, while triangular grids ensure consistent results in single runs. A two-hectare site took 43 minutes to calculate on average. This suggests that further development of the path planner could lead to significant improvements, enabling the management of sites larger than a few hundred nodes. However, the calculation time is justified for the reduced path length during deployment. The study presents a methodology that supports manual operators and establishes foundations for full autonomy.

Research on automated seedling planting systems in forestry is a crucial aspect of forestry automation. This paper introduces the development of a vision-based automated seedling planting position selection system, integrated with hardware and software components on an unmanned forest machine platform. Developed around object detection as the core, this research presents a comprehensive system consisting of two main functionalities: (i) A vision system that performs obstacle detection and localization, providing estimated obstacle types, sizes, and positions to the plant planner function. (ii) A plant planner function utilizes this information to plan the plantable areas and selects suitable planting locations. The integrated system has been tested in the field and we found it to effectively determine suitable planting locations on the ground of a clear-cut. The implementation of this system lays the foundation for subsequent automated planting operations. Furthermore, the automation of forest seedling planting reduces the need for manual labor and enhances planting precision, contributing to improved forest health and ecological balance. Looking ahead, this research offers insights into the future development of unmanned forestry operations, making strides in automating forest management, achieving cost-effectiveness, and facilitating ecological restoration.