The demand for autonomous functions in forestry operations is growing due to labor shortages, safety risks, and the need for precision in unstructured forest environments. However, dense vegetation, occlusions, and unreliable GNSS signals make autonomous navigation challenging for forest machinery. To address these issues, this paper proposes a LiDAR-based tree detection and parameterization method to support Simultaneous Localization and Mapping (SLAM), which is essential for reliable long-term autonomy. The proposed approach consists of three components. First, density-based spatial clustering segments tree-like structures from raw LiDAR point clouds. Then, parametric modeling fits cylindrical representations to trunks using RANSAC and numerically stable circle fitting. Finally, validation constraints, including a radius consistency check for maintaining stable trunk geometry and a probabilistic visibility filter for excluding occluded or back-facing points, enhance robustness against noise and occlusion-induced outliers. The method is evaluated in controlled 2D LiDAR simulations with realistic noise and branch interference. It achieves an RMSE of 0.077 m for trunk radius and 0.062 m for center localization compared to the simulated tree radius of 0.27 m. These results demonstrate the method's ability to extract typical tree landmarks from single scans, even under challenging conditions. By enabling accurate tree modeling in cluttered forest environments, the proposed method lays a foundation for incorporating structured environmental features into SLAM pipelines.

The automation in the construction machine field requires a robust understanding of their surroundings and should be able to localize and classify surrounding objects robustly. State-of-the-art object detection algorithms are usually deep learning-based approaches that take red-green-blue (RGB) images as the only input. However, recent findings highlighted the limitation that these deep learning-based approaches may not perform robustly due to the bias introduced by the training set, which is also a common problem in existing construction machine data sets. As a result, the object detection performance on out-of-distribution (OOD) data is significantly worse than on the training set. This may cause severe accidents and unexpected economic losses on smart construction sites. To address this issue, this study proposes a novel object detection algorithm, called "Temporal- and Appearance-Guided Object Detection" (TAG), which optimally extracts information from temporal information (optical flow) and appearance information (RGB) to improve object detection accuracy and robustness despite OOD data. To evaluate the performance on various OOD data sets, a custom construction machine data set generation system is created that enables nonoverlapping training and testing distributions. Tests with the simulated data set and the real-world data set are performed considering the diversity of the working conditions and the challenge of the OOD data. Compared with existing typical alternative solutions, the results show strong empirical evidence that the proposed construction machine object detection algorithm significantly increases the robustness and generalization capability in dynamic cases without compromising performance in static cases.

—Electrification is a promising trend for wheel loaders, with their advantages of high efficiency, zero emissions, and low noise. However, wheel loaders experience high instantaneous power and frequent acceleration and deceleration during operation, accelerating battery aging and increasing the annual operating cost. One approach to extending battery lifetime is to employ an electro-hydraulic hybrid drivetrain system to reduce battery usage frequency and charging/discharging currents. In this paper, a cycle-adaptive control strategy is proposed with the combined goal of minimizing electricity and battery aging costs. Dynamic programming is used to determine the optimal power ratio for different speed trajectories. Based on the offline results, the cycle-adaptive control strategy is designed to adjust power distribution based on vehicle position, speed, and average hydraulic pressure of the last cycle. Simulation results indicate that the proposed strategy achieves combined costs comparable to those of dynamic programming and reduces costs by approximately 7% compared to the universal thermostat control algorithm. Additionally, the paper discusses the impact of electricity prices in different countries and battery prices over various years on electricity consumption and battery degradation, providing design guidelines for implementing the strategy in practice.

Electrohydraulic actuators (EHAs) are critical in applications demanding compact designs, high power density, and precise motion control. However, their nonlinear dynamics and inherent uncertainties pose considerable control challenges. This study develops a reinforcement learning (RL)-based control framework, integrating an actor-critic algorithm [deep deterministic policy gradient (DDPG)] and a novel reward-shaping method to address these challenges without requiring prior expert knowledge. The proposed approach is validated on a heavy-duty EHA-driven inverted pendulum testbench, as a benchmark system for nonlinear and unstable dynamics. Experimental results demonstrate the RL controller's effectiveness in achieving the dual-control goal: swing-up and balancing of the EHA-driven pendulum. Furthermore, a comparative analysis with the classical linear quadratic regulator (LQR) highlights the strengths and limitations between RL-based control and model-based control. This study serves as fundamental research, offering practical and theoretical contributions to the application of RL in advanced motion control for EHAs and other complex systems.

Increasing the process efficiency of agricultural tasks is a key measure to decrease overall costs and CO2 emissions. However, optimizing tractor–implement combinations is challenging due to the variety of processes and implements and the complexity of the powertrains in modern tractors. In addition, overall process efficiency is an ambiguous optimization objective in agricultural processes as it relates resource consumption to harvest yields, which are only known at the end of a harvest season. The presented approach defines process constraints, ensuring optimization does not negatively affect harvest yield. These constraints allow for the formulation of explicit objective functions that are observable during the operation. The method establishes a mathematical foundation for the optimization of agricultural processes. The mathematical principles of the theoretical framework and the techniques used to define control constraints are explored, whereby the applicability to alternative objectives like optimizing the overall process cost is highlighted. To demonstrate the practical utility of the proposed approach, an optimization cycle is applied to a real-world scenario: adapting the working speed during the tillage process using a cultivator to maximize energy efficiency. The approach simplifies the optimization problem by formulation as a constraint optimization problem, allowing for improving the operating point of tractor–implement combinations with respect to observable process objective functions. The results underline the importance of advanced control strategies in agricultural machinery, advancing precision agriculture and promoting sustainable farming practices.

In the context of automating tillage, the driving task is largely considered solved. However, monitoring the tillage process and its associated parameters remains an open challenge. Tillage and its characteristics have a significant influence on plant germination and growth. As a result, understanding these parameters is crucial for optimizing crop yield and, by extension, the agri-food sector. This study presents a novel approach to measuring the aggregate size distribution using camera images and machine learning algorithms. Additionally, it investigates the impact of diameter estimation on the parameter metric. The proposed algorithms achieve recall values of up to 0.731 on segmentation masks in the test dataset. This method enables the large-scale, detailed analysis of soil aggregates, contributing to a better understanding of how soil structure affects field emergence and crop yield, thereby supporting the broader goal of automation in tillage.

Intelligent control systems are a promising approach for optimizing and automating agricultural machinery. For these algorithms, precise knowledge of the machine’s characteristics and its interaction with the environment is required. However, many input variables required to describe the machine’s context are not easily measurable during operation, making process-specific modeling challenging. This paper presents a novel approach utilizing a neural network-based encoder to extract information from spatially close reference measurements. The reference measurements are artificially compressed into a low-dimensional latent representation. The information is then used to adaptively model machine states without the need for explicit measurement of all influences. By leveraging the spatial correlations, the method enhances the predictive accuracy of machine states by accounting for non-directly measurable situational effects. The experimental results demonstrate the effectiveness of the method. The proposed modeling method enables control algorithms to adapt to situational influences and contributes to the automation and optimization of agricultural field work.

Automation of mobile machinery is critical in the construction industry to improve efficiency and ensure safety. Perception technologies, particularly for detecting and monitoring the actions of construction machinery, are essential for optimizing workflows and mitigating accident risks. However, the complex nature of construction environments, the variety of machines, and the dynamic interactions at construction sites pose significant challenges for reliable object detection and action recognition. This study introduces a deep learning approach using temporal vision information for object detection and action recognition of mobile machinery in construction environments. In particular, a novel strategy called Integrated YL-SF is proposed, which integrates the YOLOv8 framework with the SlowFast model enhanced by Transformers to achieve robust action recognition and motion analysis of construction machinery. The proposed method is evaluated on a custom dataset with a variety of machine types and real-world operating environments, and it is benchmarked against the standard YOLOv8 model. The results show that the Integrated YL-SF framework outperforms existing methods and effectively addresses challenges such as dynamic scenarios, object occlusion, and multi-machine interactions in complex environments.

The automation of hydraulic excavators is significant for enhancing productivity and safety in uncertain and dynamic environments. Achieving autonomous operation requires advanced control strategies capable of handling system constraints, nonlinear hydraulic dynamics, and complex environmental interactions. This study proposes a reinforcement learning (RL)-based methodology to perform a complete excavation cycle by controlling proportional valves. A comprehensive joint simulation tool is developed, in which a hydraulic system model is detailed based on a real machine, and it is integrated with an excavator mechanism and working environment to create a realistic interaction environment for RL training. The RL agent, trained using Proximal Policy Optimization (PPO), incorporates a customized reward shaping method that ensures operational safety and accuracy, considering constraints such as pump flow saturation and geometric constraints. In addition, an Adaptive Control Frequency (ACF) method is developed to enhance training efficiency by dynamically adjusting the control frequency based on task complexity. Comparative validations demonstrate the RL agent’s ability to successfully complete a full excavation cycle, satisfy operational constraints, and generalize across varying initial conditions and valve responses. Furthermore, the controller operates effectively in a soil environment despite being trained without soil, demonstrating robustness to uncertain, time-varying loads.

This paper proposes an automatic method for excavator working cycle recognition using supervised classification methods and motion information obtained from four inertial measurement units (IMUs) attached to moving parts of an excavator. Monitoring and analyzing tasks that have been performed by heavy-duty mobile machines (HDMMs) are significantly required to assist management teams in productivity and progress monitoring, efficient resource allocation, and scheduling. Nevertheless, traditional methods depend on human observations, which are costly, time-consuming, and error-prone. There is a lack of a method to automatically detect excavator major activities. In this paper, a data-driven method is presented to identify excavator activities, including loading, trenching, grading, and idling, using motion information, such as angular velocities and joint angles, obtained from moving parts, including swing body, boom, arm, and bucket. Firstly, a dataset lasting 3 h is collected using a medium-rated excavator. One experienced and one inexperienced operator performed tasks under different working conditions, such as different types of material, swing angle, digging depth, and weather conditions. Four classification methods, including support vector machine (SVM), k-nearest neighbor (KNN), decision tree (DT), and naive Bayes, are off-line trained. The results show that the proposed method can effectively identify excavator working cycles with a high accuracy of 99

By leveraging data from both RADAR and LiDAR sensors, the accuracy of object detection and other autonomous driving tasks significantly improves in comparison to single-sensor approaches. This paper introduces a novel adaptation of the low-level fusion variant of Complex-YOLO, specifically designed to cope with sensor disturbances. We develop and implement an enhanced training methodology that incorporates both functional sensor data and simulated sensor disturbances, allowing the network to maintain high performance even under data perturbations. Our empirical results demonstrate that this approach enables Complex-YOLO to effectively adapt to such disturbances, with an improved mean performance by 109% compared to a network trained only on undisturbed data. Key contributions include a robust training framework that integrates disturbance simulation directly into the training loop, significantly enhancing the detector's resilience in challenging environments, and an evaluation concept that quantitatively measures this improvement.

Researches in improving the control performance for high accuracy under uncertainties for electro-hydrostatic actuator (EHA) have never stopped in many industrial applications. In order to achieve high precision position control with high robustness, this paper proposes a novel EHA based on pump controlled hydraulic motor with a high response control scheme. First, the composition and the mathematical model with open loop characteristic of the system are introduced, the uncertain external load is established through the combination of hydraulic system with an inverted pendulum. Second, in order to achieve position control under the internal and external uncertainties, this paper proposes a control scheme based on Fuzzy PID and ANFIS, while the former handles the load uncertainties transferred from the pendulum and the latter achieves the compensation of the output flow rate from the hydraulic pump in order to acquire more precisely position control. Finally, to verify the control accuracy with robustness verification, an instantaneous disturbance acting on the pendulum rod is given, with the comparison of simulation results, the supercity of the proposed method compared to traditional controller is verified.

Excavators are crucial in the construction industry, and developing autonomous excavator systems is vital for enhancing productivity and reducing the reliance on manual labor. Accurate estimation of the volume of the excavator bucket fill is key for monitoring and evaluating system automation performance. This paper presents the use of 2D depth maps as input to a Faster Region Convolutional Neural Network (Faster R-CNN) deep learning model for bucket volume estimation. This structure enables high estimation accuracy while maintaining fast processing speed. An excavator operation monitoring test bench was established, and the datasets used in the study were self-generated for training. A loss function is proposed, combining Cross Entropy with Root Mean Squared Error to improve generalization and precision. Comparative results indicate that the proposed approach achieves 96.91% accuracy in fill factor estimation and predicts in real-time at about 10 fps, highlighting its potential for practical use in automated excavator operations.

This paper investigates the impact of training data with different quantities and compositions on the performance and robustness of a Neural Network (NN) controller for the wheel loader bucket filling task. Collecting training data for machine learning methods with a real-world Heavy Duty Mobile Machine (HDMM) is expensive, and therefore knowing how to collect the data and in what quantities will significantly reduce the data collection effort. We collected 2000 bucket fillings of non-homogeneous material, more specifically, a blasted rock pile with a kernel size of 0-400 mm. No previous study has reported such a challenging material composition. The collected data was divided into 6 datasets with sizes of 10, 20, 50, 100, 500, and 2000 bucket fillings. We use the Dynamic Time Warp (DTW) distance, k-medoids clustering, and the silhouette score, to create diverse and dissimilar datasets. Furthermore, one additional dataset was created with 10 bucket fillings, which are as similar as possible, resulting in 7 datasets in total. The datasets were used to synthesize 7 controllers that were then evaluated with a set of experiments to compare their performance to one another and the human operator. The results showed that the controller trained on similar bucket fillings was not robust and had poor performance, as expected. The experiment also showed that all the controllers trained on diverse data were robust enough to load the blasted rock material. However, the loaded material weight was less than the human operator, where the best controller loaded 9% less material weight, but 11% faster than the human operator.

Loading multiple different materials with wheel loaders is a challenging task because various materials require different loading techniques. It’s, therefore, difficult to find a single controller capable of handling them all. One solution is to use a base controller and fine-tune it for different materials. Reinforcement Learning (RL) automates this process without the need for collecting additional human-annotated data. We investigated the feasibility of this approach using a full-size 24-tonnes wheel loader in the real world and demonstrated that it’s possible to fine-tune a neural network controller that was originally trained with imitation learning on blasted rock for use with an unknown gravel material, requiring 20 bucket fillings. Additionally, we showcased the adaptability of a controller pre-trained on woodchips for an unknown gravel material, requiring 40 bucket fillings. We also proposed a novel reward function for the material loading task. Finally, we examined how the sampling time of the reinforcement learning algorithm affects convergence speed and adaptability. Our results demonstrate that it’s optimal to match the sampling time of the RL algorithm to the delays of the wheel loader’s hydraulic actuators.

Hydraulic mobile machines in the construction industry, such as excavators, rely on hydraulic actuators to handle various workloads. A reliable and accurate classification of these workloads is essential for improving the safety and automation level of machine operations. Processing of hydraulic actuator operating data via a deep-learning-based classifier has great potential for time-varying workload recognition. In particular, Long Short-Term Memory (LSTM) networks demonstrate strengths in analyzing time-sequential information. In this study, a customized LSTM-based classifier is developed to perceive the uncertain and varying external workload in real-time accurately. Considering the working features of the bucket cylinder, signals including position, velocity, and pressure from both chambers of the hydraulic cylinder actuator are selected as input data for the classifier. An experimental dataset from a hydraulic actuator test bench, reflecting typical workload types of excavator digging processes, is collected. The proposed classifier achieves satisfactory accuracy in recognizing workload types in both training and test cases, demonstrating its practical advantages for excavator digging processes.

Reinforcement Learning (RL) requires many interactions with the environment to converge to an optimal strategy, which makes it unfeasible to apply to wheel loaders and the bucket filling problem without using simulators. However, it is difficult to model the pile dynamics in the simulator because of unknown parameters, which results in poor transferability from the simulation to the real environment. Instead, this paper uses world models, serving as a fast surrogate simulator, creating a dream environment where a reinforcement learning (RL) agent explores and optimizes its bucket-filling behavior. The trained agent is then deployed on a full-size wheel loader without modifications, demonstrating its ability to outperform the previous benchmark controller, which was synthesized using imitation learning. Additionally, the same performance was observed as that of a controller pre-trained with imitation learning and optimized on the test pile using RL.