Learning Curves of Harvester Operators in a Simulator Environment

Abstract

Simulator training helps provide safe and cost-effective training for operators of modern forestry machines that require high motor skills, constant concentration, and proper planning. The aim of the study was to analyze the learning curves of the trainees in order to determine the period during which most development takes place. In this study, 11 trainees were trained on a John Deere harvester simulator for approximately 15 h each. In each case, a clear learning curve could be identified, despite high inter- and intra-person variability. Effective time showed a steady decrease during training, with a group minimum at the end of training (1.25 min). Crane tip distance per tree dropped rapidly in the first 3–4 h, followed by a more gradual decrease to reach a minimum of 23.8 m. Crane control showed a significant increase from an initial 0.63 to a maximum of 0.8 by the 9th hour of training. A number of crane functions used simultaneously increased more rapidly to almost a maximum value (1.8) already in the 5th hour. The individual curves for each trainee were highly variable, showing a wide range of values and shapes. In conclusion, most personal development occurs during the first phase of simulator training, which typically takes approximately 9–10 h. It is important to consider significant inter-personal variability and tailor the duration of simulator training to individual needs.

1. Introduction

Modern forestry harvesters are the dominant machines for felling and processing trees in professional forestry in Europe. These machines are highly efficient but also very complex, which places significant pressure on the operator and their capabilities [1,2,3,4,5,6]. The operator’s innate abilities and the skills acquired during training are crucial for achieving good results and the expected productivity [7,8,9]. The competence requirements are therefore high, but the formal training of future operators is still not well established in many countries [10,11,12]. Traditionally, new operators are trained “on-the-job” by more experienced colleagues [11,13]. In this scenario, harvester operators usually start by gaining experience on a forwarder, which helps them get used to driving in rough terrain and to joystick and crane operation. They also become familiar with the timber assortments and general working conditions in the forest. Only then are they promoted to work on the harvester, which requires some new skills (interacting with a computerized measurement system and making decisions on bucking) and is more demanding in terms of the required speed of crane movements [1,11]. However, this traditional method can have serious drawbacks: it is expensive, risky, and tends to pass on bad habits. All these disadvantages could be alleviated using simulator training. Not only is it cheaper and completely safe for both the operator and the environment [14,15,16,17,18], but it also often provides guidance on how to perform basic tasks. One example of such a software guide is John Deere’s TimberSkills suite. It consists of a multitude of structured tasks that help teachers and students build skills in a controlled manner. The description of any particular task advises on the most effective working technique, provides some goals to be achieved in order to pass, and is often accompanied by an instructional video. This provides an objective training standard and assists in the personal development of the prospective operator. Some of the drawbacks of simulator technology are: limited access to the equipment for companies looking to train a single operator; and some discrepancies between virtual reality and real-world conditions [11,13,16,19,20]. Another problem is the lack of personnel trained and experienced in using simulators as a teaching tool, as well as proven and tested training curriculums [17].

Simulators are widely used for training in many professions: air, wheel, and rail transportation [21,22,23], construction [14], medicine [24], military [25], and many others where complicated and expensive equipment is used or where the consequences of a beginner’s mistake would be dangerous. Logging equipment falls into both of these categories, and simulator training of logging machine operators is widely used, especially in Scandinavia [13,15,16,26]. Logging simulators can also be used for other purposes, such as selecting the most promising operators [11,12,18,27] or for scientific research [5,28,29,30].

A learning curve could be defined as “the rate of someone’s progress in learning a new skill”. It could also be described as the relationship between productivity and experience, or the improvement of performance through learning over time [27,31,32,33]. In forestry, productivity or time consumption are typically the preferred measures of performance [34]. The learning curve can be characterized as a relationship between productivity and time per work cycle [31]. In a simulator environment, however, other variables become available that can also describe a trainee’s improvement over time, such as crane tip distance, crane control, or the number of simultaneous boom movements. In particular, the last two measures appear to be more suitable for formally quantifying a trainee’s progress, as they are not dependent on factors such as tree size or cutting method. Simultaneous boom movements reflect an individual’s innate ability—some people have better coordination than others and can be expected to achieve higher scores on this aspect [12,18,35]. Crane control describes the decisiveness and precision in crane work and is expected to be associated more with the level of training than natural talent. With experience, some of the operator’s movements are combined into “automatic” sequences that are controlled as a single “chunk” by the cerebellum rather than the cerebrum [31]. This effect is expected to improve performance in both simultaneous boom movements and crane control, and while it is learned in the simulator environment, it can potentially be used in real work.

Limited research exists on the effectiveness of simulator training for forestry machine operators and the extent to which skills acquired in a simulator environment transfer to real-world scenarios [11,17,27,36].

Another important consideration in operator training is determining the appropriate period required for acquiring basic skills. The time frame varies from person to person based on their individual aptitudes and abilities, so it should be estimated as a range. However, any structured training program must recognize personal variability and provide sufficient practice time for most individuals to gain adequate experience. It is common for progress to occur primarily at the start of the learning process, when the learning curve is at its steepest [27]. Over time, the slope of the curve decreases until it reaches a plateau, and may even experience a slight decline, resulting in a sigmoid shape for the learning curve [31]. In a study by Gellerstedt [2], the time required to effectively use the machine is estimated to be about five years, with variations for different work tasks and operators. Purfürst [31] estimated that harvester operator performance roughly doubles within the first 8 months of training. In a South African study by Wenhlod et al. [27], the learning curves for harvesting operators were reported to reach and end after approximately 6.5 months for clearcutting and 9 months for thinning. According to a report by Calabrese, a chainsaw logger requires approximately 8 months of training to become a harvester operator [37]. Such a long training period incurs significant costs for the employer [31], but can be shortened by structured training.

A thorough understanding of operators’ learning curves, both on simulators and real machines, is essential to determining the required training period and developing training schedules suitable for most individuals. This study attempts to describe the learning curves of young trainees during simulator training, focusing not only on productivity but also on the economy of crane movements (crane tip distance), precision (crane control), and fluency (simultaneous moves).

Individual differences were analyzed, and variability was taken into account. The results could assist in designing a training curriculum that utilizes the time and effort of a trainee to the maximum while recognizing their personal talents and abilities. Early recognition of a particular student’s innate skills will help tailor the training content and schedule to produce the best results.

2. Materials and Methods

Data for analysis were collected in July 2021 during a harvester operator course for final year students of the Forestry High School in Zagnansk (southern Poland), conducted by an external company (Forest Consulting Center Sp. z o.o.). The two-week course comprised 10 days of simulator exercises mixed with theoretical classes, and later practical exercises in the forest. The group of students participating in the course consisted of 10 males and 1 female, 19 years old, with no previous experience in crane work or mechanical harvesting. The student group was selected by the school authorities, and the course was intended to be a reward for achieving top grades. The students are referred to by their initials throughout the manuscript. Simulator training consisted of completing tasks from the comprehensive TimberSkills training software on a John Deere harvester simulator. After each 30 min training session, students were required to complete a standardized assessment task. A training period of 30 min was selected based on the average duration of TimberSkills training, which is approximately 10 to 15 min. This allowed the students to complete two or three exercises between assessment tasks. A standardized assessment task was designed to be brief and straightforward, with the objective of enabling even inexperienced trainees to complete it within a maximum of three minutes. The results of these attempts were stored and subsequently analyzed. Over the course of the training, students completed approximately 15 h of simulation practice with 28 repetitions of the assessment task. This task was adapted from one of the TimberSkills trainings and consisted of felling two small trees (average volume 0.34 m³) and processing the logs into assigned assortments, placed in a marked area (6 logs per trunk), without the need to drive the machine (although it was allowed). The variables used in the analysis were extracted from the scoring report in a PDF file and exported to MSExcel using PowerQuery, and later exported to RStudio for statistical tests in R (version 4.2.1) [38]. The simulator reports allow analysis of many factors describing the trainee’s work, but for the most informative learning curve, a number of variables were selected: effective time (which corresponds to productivity as the procured volume was always the same), crane tip distance per tree, crane control, average simultaneous boom movements, and heap quality. In addition, an analysis was made of the groups of top three and bottom three trainees at the beginning and end of the course with respect to the variables mentioned above. For this analysis, we used the average of the first three attempts and the average of the last three attempts to reduce variability, particularly at the beginning.

Effective time was obtained by subtracting idle time from total time and was expressed in minutes. It was treated as an approximate measure of productivity, since all students had to fell and process the same two trees, and the volume produced was the same. The term “crane tip distance” refers to the length of the path that the crane tip travels in 3D space and is measured in meters. In addition to the total tip distance, the simulator reports also provide the tip distance per tree. This is useful for comparing exercises with different numbers of trees to fell. In this study, this measure was used even though the number of trees was always the same. Crane control is a parameter that takes values from 0 to 1 and reflects the decisiveness and precision of the crane movements. This variable was estimated by an internal algorithm of the simulator software, where smooth movements and placing the harvesting head on the target on the first attempt were rewarded, while shaky, imprecise movements and multiple attempts to place the head in the correct position caused a reduction in the score. Simultaneous boom movements are the average number of crane functions used at the same time, as described by the number of hydraulic cylinders in action at the same moment. Activating many functions simultaneously requires increased mental effort and is usually a sign of experience or innate ability.

Two variables related to processed wood piles were “Heap Quality” and “Heap Size”. The first is a parameter determined by the simulator software factoring in whether the logs are laid parallel, close together, and with their tops in a line, and takes values from 0 to 1. The second is the average number of logs in a pile. A skilled operator leaves neat stacks that are easy for a forwarder to pick up and are about the size of a full grapple load. The harvester’s working time was divided into different operations: crane work, wood processing, sawing, feeding—these times were recorded separately. The sum of these times did not necessarily equal the total effective time, because overlapping operations would be measured many times. Another measure of the quality of the harvester operator’s work is the amount of damage caused to the machine and the environment. The simulator report distinguishes between machine damage (divided into machine, cab, crane, and saw damage) and tree damage (“boom-to-tree”). Damage to the remaining trees was only counted if the tree was not later felled. Finally, the correct height of the stump was assessed; the report shows the number of stumps between 20 and 30 cm and over 30 cm.

Due to the irregular and noisy nature of the data, the normality assumption was often not met (tested with the Shapiro–Wilk normality test), and the test used was the nonparametric Friedman ANOVA with Durbin–Conover post hoc test from the ggstatsplot package [39]. The assumed significance level was α = 0.05.

3. Results

The simulator reports provided data that enabled the analysis of student performance across several variables, including some that were previously unreachable, such as boom tip distance, crane control, and simultaneous boom movements. The traditional metric of productivity has been replaced by task execution effective time, as the task was uniform for all attempts.

3.1. Effective Time

During the training, students were advised against prioritizing high productivity levels as it can result in increased stress and a hectic working style. The score report does not provide a direct reading on productivity in m³/h, although the time taken to cut and process the same two trees serves as a productivity measure.

The average task completion time for all trainees and attempts was 1.57 min (1 min and 34 s, SD 0.36 min), with the slowest time being 2.84 min (PW in the 1st attempt) and the fastest time being 0.99 min (KW in the 9th attempt). For individual students, the lowest mean completion time was achieved by operator JK at 1.26 min, followed by AS at 1.31 min and KW at 1.35 min. The longest individual average completion times were recorded for PW, HB, and DK, at 1.92, 1.78, and 1.77 min, respectively.

The Friedman ANOVA test for the whole group revealed statistically significant differences in effective time among attempts (χ²Friedman(27) = 110.33, p = 4.89 × 10⁻¹²), with Kendall’s W (0.37) indicating “fair agreement” [40]. Pairwise comparisons showed significant differences between the initial stages (attempts 1–3) and later stages (attempts 16–28), as well as between attempts 4 and 8 in comparison to attempts 23 and 28. The median time of the group decreased from 2.09 min in the first attempt to 1.44 min in the final attempt (28), with the lowest time recorded in the 27th attempt (1.30 min), as shown in Figure 1.

For each trainee, the completion time always decreased (Figure 2). This reduction often starts with a sharp decrease during the initial phase (most apparent in the cases of AZ, OS, PW, and JK) or a steady downward slope (AS, DK, HB, and HJ). In some cases, there was an incline during the final stage (AS, HB, HJ, and JK). In some cases, the curve displays a distinct pattern of sharp decline followed by a plateau period, during which there is minimal improvement (AZ, JK, OS). Typically, the fastest operators at the beginning of the trials remained the fastest at the end (AS and KW), as did the slowest (HB and PW).

3.2. Boom Tip Distance

Distance traveled by the tip of a harvester crane is an effective indicator of an operator’s proficiency, particularly reflecting proper planning of work, although not necessarily reflecting their manual dexterity. Minimizing unnecessary crane movements increases efficiency, reduces the risk of damage to remaining trees, and decreases wear on machine components.

On average, the crane tip traveled 26.8 m per tree (SD = 4.52). KW achieved the shortest distance of 16.82 m on the 18th attempt, and HB moved the crane the longest distance of 44.38 m on the first attempt. Throughout the course, JK was the most economical with crane movements, with an average distance of 22.7 m, while the longest average distance was obtained by HJ (30.1 m). In the initial phase of the training, the crane traveled the longest distance (36.3 m on average on the first attempt), with a significant drop for 6–7 attempts, and then stabilized, reaching the minimum group average on the 28th attempt (23.5 m). This was confirmed with statistical tests (χ²Friedman(27) = 124.90, p = 1.53 × 10⁻¹⁴), where pairwise comparisons showed significant differences between attempts 1 and 4 and most attempts above 9, attempts 4–8 with some above 15, and no differences above attempt 9. The median crane distance for the group decreased from 37.7 m in the first attempt to 23.8 m in the 16th and final attempt (28), as shown in Figure 3.

Overall, the students’ crane movement economy improved throughout the course, with each student finishing with a shorter distance than at the beginning (see Figure 4). Some students improved at a fairly constant rate (HJ, HW), while others experienced a large initial decrease in distance followed by a much smaller decrease (DK, HB, JK) or even a slight increase (AZ, KW, OS, and SD). Of the top three at the beginning (AS, AZ, and OS), only one remained in the top three at the end (OS), while the best at the beginning (AS) was in the bottom three at the end. Furthermore, one of the bottom three at the beginning—DK—got into the top three at the end, while the others (HJ and PW) kept their lowest positions.

3.3. Crane Control

Crane control is an index given by the simulator algorithm (between 0 and 1) describing the precision and confidence of the crane movements. Any hesitation, corrections, or missing the target were penalized by lowering the score.

The mean crane control score for all trainees and attempts was 0.729 (SD = 0.078), with scores ranging from 0.47 (PW in the fourth attempt and SD in the first) to 0.88 for AS in the 23rd trial. Among the students, AS achieved the highest crane control score with a mean of 0.801 and HJ the lowest with a mean of 0.673. The highest group median for crane control (0.8) was achieved in the 17th attempt, followed by 0.79 in the 14th, 15th, and 20th attempts, while the highest group mean was achieved in the 14th (0.773), 11th (0.767), and 16th (0.766) attempts.

The Friedman ANOVA test revealed some significant differences in crane control with increasing experience (χ²Friedman(27) = 82.59, p = 1.52 × 10⁻⁷), but pairwise comparisons showed only a limited number of different pairs. Attempts 1 and 2 were significantly worse than 11, 14, 16, 17, 20, and 28, and no attempt above 4 showed significant improvement (Figure 5).

The individual learning curves show some common patterns—in most cases, an initial steep increase is followed by a peak and often a decline (Figure 6: AS, JK, KW, PW, SD). Conversely, HJ and HW exhibited only a slight overall improvement in crane control (0.057 and 0.043, respectively), while PW demonstrated a slight decrease (by 0.003). Two of the top three initial scorers remained in the top three at the end of the course (AS and OS), and only one of the lowest initial scorers remained in the bottom three (PW).

3.4. Simultaneous Crane Movements

In addition to confidence and precision in crane movements, a very useful metric describing the operator’s skill is the number of crane functions used at the same time. The more simultaneous movements, the smoother and more effective the crane operation. This variable is easily accessible in the simulator environment, whereas in the real world it is extremely difficult or even impossible to determine.

The overall average for all students and all attempts was 1.73 crane functions used simultaneously (SD = 0.239), ranging from 1.14 (SD in the second trial) to 2.47 (AS in the 19th trial). These two students also achieved the highest (AS-2.10) and lowest (SD-1.38) individual average number of simultaneous crane movements throughout the course. Similar to the crane control, the highest group median for simultaneous movements (1.82) was achieved in the 20th attempt, while 1.8 was already reached in the 10th, 15th, and 16th, and the highest group mean was reached in the 16th (1.86), 21st, and 20th (1.85) attempts (Figure 7).

Partially similar to the crane control, statistical testing revealed some differences (χ²Friedman(27) = 93.84, p = 2.58 × 10⁻⁹), and pairwise comparisons indicated the first 4 attempts differed significantly with most attempts above the 13th.

Even more pronounced than for the crane control, the individual learning curves followed a similar pattern—steep initial slope, peak, plateau, or decline (all except DK, who improved fairly steadily) (see Figure 8). Of the top three performers at the beginning, only one was in the top three at the end (AS), while of the bottom three at the beginning, two students remained in that group (HJ and SD). Nevertheless, one of the bottom three at the beginning (AZ) improved so much that he was the second-best performer at the end in terms of simultaneous crane movements. This was the largest increase (from 1.38 to 2.05) recorded for any student, while the smallest was for JK, who finished the course with the same score as at the beginning (1.77).

3.5. Other Variables

The neat and organized piles of processed wood left orderly on the ground are a good indicator of the harvester operator’s proficiency. The variable “Heap Quality” (index with values from 0 to 1) refers to the proper alignment of the log tops (more or less even) and the orientation of the logs (parallel to each other)—in a way that facilitates lifting with the forwarder’s grapple. The variable “Heap Size” refers to the number of logs in a single heap—ideally forming a full grapple load.

The average heap quality index for all trials was 0.737 (SD = 0.129), and the statistical test showed no differences with increasing experience (χ²Friedman(27) = 35.39, p = 0.13). There was no noticeable pattern in the individual learning curves.

The median heap size was 3 logs (mean 3.31, SD = 1.24), ranging from 1.5 to 6. There was no clear trend for the number of logs in a pile to increase with experience, either for the whole group or for individual students.

Harvester operations were also timed separately: crane moving, processing, sawing, and feeding. The first two showed improvement similar to total effective time; attempts 1–8 were found to be different from most over 15. The average crane moving time was 1.22 min (SD = 0.331) per attempt, ranging from 2.3 for PW in the first attempt to 0.65 for JK in the 24th. Processing time averaged 0.907 min (SD = 0.203) and ranged from 1.75 for AZ in trial 2 to 0.57 for KW in trial 9. The shortest average crane movement and processing times were recorded in trial 27 (0.985 and 0.771 min, respectively), and the longest in the first attempt (crane movement 1.64 min) or in the second attempt (processing 1.17 min). Sawing time and feeding time (total averages of 0.14 and 0.312 min, respectively) remained fairly constant regardless of student experience.

One of the advantages of training in a simulator environment is the ability to accurately count the damage done to the machine (cabin, boom, saw) or to the remaining trees. In this study, however, the exercise was too short, and the number of errors made was too small to make any meaningful generalizations. Trainees made a total of 106 errors that resulted in damage (65 chainsaw damage, 21 machine damage, and 20 “boom-to-tree” damage). It should be noted that the number of damages to residual trees may have been significantly higher. However, since all trees were eventually felled, these incidents were not counted.

Another operator error is not cutting the tree low enough, resulting in high stumps. The simulator software tallies the number of stumps that are over 20 cm and over 30 cm in height. The occurrence was infrequent, with only 17 stumps measuring over 20 cm in total and none exceeding 30 cm. As a result, no trends could be identified.

4. Discussion

The general overview of the analyzed data indicates that certain scores were somewhat erratic and noisy. The results of consecutive attempts, interrupted only by approximately half an hour of training, varied considerably at times. This effect is normal, especially in the beginning phase of the learning curve, where greater variability reflects trainees’ varying levels of coordination and skills. The variability in the metrics connected with productivity, such as time of completion and crane distance, was particularly noticeable. The initial attempts were widely dispersed, while the subsequent ones tended to converge into a narrower range. This result is in line with Burk’s [11] research, which examined forwarder trainees after both traditional and simulator-based training and discovered a reduction in within-group variation over time. The variability of results may be due to differences in concentration, daily well-being, physical disposition, and even the amount of sleep [11,31].

Despite the considerable inter-person variability, overall trends were usually visible and formed distinctive learning curves. As is evident, the brief training period only permits the design of an “initial learning curve”. This is particularly pertinent in the context of simulator training, which usually takes only a few days [3,11,12,27]. For instance, Wenhold et al. [27] conducted a simulator training in South Africa to study the learning of harvester operators. The training lasted an average of 8.4 days, with approximately 25 tests per operator (three tests per day). They calculated a skills development metric using a performance level (PL). An increase in PL (calculated as the current score divided by the population mean score) indicated an improvement in the trainees’ skills. In our study, we did not include PL in the “Results” section because this metric is strongly dependent on a particular group of trainees (population mean), making the results incomparable. However, we calculated PL regarding the task completion time solely for the purpose of this reference. Our results closely reflect those obtained in South Africa, where the initial PL for clear-fell operation was 0.45 and increased to 1.77 over an average of 5.2 days (15.6 tests on average). In our study, the average minimal PL for each trainee was 0.5, and the maximum PL was 1.28 on average. On average, there was a 193% personal improvement in PL, with AS or JK gaining 75% and PW gaining 533%. As shown in Figure 2, AS and JK were among the fastest throughout the course, resulting in little improvement, but PW was the slowest, both at the beginning and end, despite outstanding improvement in PL. This is in line with the findings of the subsequent phase of operator training in South Africa, during which trainees operated actual machinery [27]. The operator who began with the lowest PL experienced the greatest PL gain, increasing from 0.3 to 1.15 by the end of the training. Conversely, the operator who performed the best from the outset (initially scoring 0.62) had the smallest increase, reaching 1.5 by the end of the training, a 141% increase. A similar study was conducted in Brazil, which involved simulator training for 12 young trainees and produced comparable results [12]. The task completion time showed the most significant improvement during the first four hours of training. Furthermore, the trainees who were the fastest and slowest remained in their respective categories upon completion of the training. Another study that analyzed task completion time for forwarder operators found that both time and variability decrease with experience. This finding was confirmed in our results [11].

Although time of task completion is a commonly used metric for performance on simulators, it heavily relies on factors such as the number of trees to cut, tree size, cutting type, and assortments. While the digital environment of a simulator allows for the use of more sophisticated indicators of operator performance, these are rarely utilized in published studies [11,12,16,28]. Measuring boom tip distance in nature was previously unfeasible. However, with the introduction of cranes with computer-assisted movements, such as Intelligent Boom Control (John Deere), it has become possible. The hydraulic cylinder sensors enable the computer system to constantly monitor the crane tip’s 3D position and calculate the distance it traveled. In the simulator, this calculation is performed directly within the computer simulation. This metric is essential for evaluating the efficiency of the crane movement. Experienced operators can plan their work to minimize unnecessary crane travel, increasing productivity and reducing the risk of hitting residual trees and damaging the stand and machine [30]. Figure 3 illustrates how the tip distance decreased during the first half of the training period, from an initial group median of 37.7 m per tree to approximately 25 m, and stabilized at this level. The individual curves generally followed a pattern of improvement, with every trainee achieving better results at the end. However, there were variations in the shape of the curves. Some curves decreased steadily (HJ, HW), while others decreased only slightly (AS, AZ). Most curves dropped sharply at the beginning and then stabilized or even slightly increased. Notably, only one of the top three trainees at the beginning remained in the top three at the end, while one fell to the bottom three (AS), and one of the initial bottom three trainees made it into the top three at the end (DK). This suggests that planning is a learned skill, and innate abilities do not guarantee a good score if there is not enough focus on crane movement economy. Trainee AS, who performed the best on the first try, did not improve much and was soon overtaken by his colleagues. Meanwhile, trainees JK and DK put in more effort at the beginning and remained the leaders from around the 7th or 8th attempt. Typically, most development occurs within the first 10 h of training. An increase in the distance traveled by the crane tip may suggest a shift in concentration towards other areas (like smoothness and speed of movement) and a slightly worse performance.

Crane control is a metric calculated by simulator software that describes the quality of crane work in terms of decisiveness, precision, and confidence. It takes values from 0 to 1. The majority of skill development occurs within the first 5–6 attempts, after which the results stabilize or even decrease. This is consistent with the findings of [12,36], which indicate that the majority of improvement occurs within the first four hours of training. However, some trainees did not exhibit any visible improvement (HJ, HW), and others experienced a decline in scores (HB, PW). The authors attribute this to a shift in focus and a decrease in concentration on this aspect of the work.

Somewhat similar metric supplied by simulator software is a number of crane functions used at the same time—"simultaneous boom movements”. Using multiple crane functions concurrently is a sign of a skilled operator, but it requires excellent coordination and manual dexterity [12]. Similarly to the “crane control” metric, overall group curve increased, but more gradually, and statistical differences were found between the first four attempts and the last ten. The individual learning curves (Figure 8) exhibited similar shapes, albeit slightly flatter, to those for crane control. This emphasizes the fact that these parameters describe similar motor skills. Also notable were the individual differences among the trainees. Trainee AS dominated the group in both “crane control” and “simultaneous boom movements” from the beginning to the end, while operators HJ and SD ranked near the bottom in both aspects. In contrast, trainee AZ showed significant improvement in both areas, moving from lower positions to the top three by the end of training.

Other parameters that were analyzed from simulator reports did not provide conclusive evidence for any generalizations. The parameters “heap size” and “heap quality” did not show any trends with growing experience. This lack of trend may be due to the small sample size of only two trees in the assessment task. The variable “cutting height” analyzed by [12,18] was described as qualitative and did not achieve statistical significance within their study. In our research, the simulator reports included the incidence of stumps cut above 20 cm and stumps above 30 cm. Such stumps are considered too high and are widely regarded as operator errors. Despite the initial training phase, such errors were rare and random, making it difficult to draw clear conclusions. The same was true for other recorded errors, such as machine damage or damage to residual trees. However, the simulator algorithm does not account for damage to trees that are eventually cut down. Therefore, since only two trees were felled, the low number of mistakes could be misleading. Schwegman et al. [18] studied the number of machine damage incidents and found an interesting distinction between the group that scored the highest in the Vienna Test System and the rest of the trainees. At the end of training, the average number of damage incidents for the high-scoring group was 14, while the average for the rest was 50. The findings of our study could not confirm this, as the total number of errors made by students was too small—only 106 in total.

The learning curve for operators can be viewed as a period of continuous performance improvement until a plateau is reached, at which point performance remains relatively constant [31,33]. This period of time is typically measured in months or machine hours. The cited time ranges vary from 3 months [17], 8 months [37], 6–11 months [31], 6–12 months [27], up to five years [2]. In our study, we were only able to describe the “initial” learning curve due to the limited training period. The learning curve for the entire group follows the sigmoid shape described by Purfürst [31], with the distinction that the initial training period of 2–3 attempts resulted in much steeper improvement. The observed difference may be attributed to the shorter time span of the simulator study. It is likely that long-term performance improvement will follow a sigmoid curve. It is important to acknowledge that trainees would not be able to reach a plateau corresponding to the “experienced” or “working phase” as mentioned by [31] in such a short period of time. It should also be noted that individual curves were more diverse in both shape and value, which demonstrates the human influence on the learning process [1,7,12,27]. Differences in performance and learning outcomes due to the personal traits of the trainee were stated in other research [12,27,35,41].

While simulator training is widely accepted as advantageous, there remains the issue of whether the skills acquired through this method can be transferred to real-world work sites. Only a limited number of studies have addressed this aspect of training. With the increasing development of technology, the quality of 3D imaging and the use of VR goggles have made simulated environments more closely resemble reality. This is important for effective skill transfer [11,17,20]. Burk et al. [11] conducted a study comparing groups trained initially on simulators to those trained solely on real machines. The results showed no significant difference in performance for skidders, but simulator-trained operators had a slight advantage over forwarders. The complexity of the machine controls may be the reason for the observed differences, according to the authors. It is possible that training on the harvester simulator could result in even greater improvements, as the controls of the harvester are much more complex than those of the forwarder. This is supported by the findings of [13], which showed that the addition of 25 h of simulator training significantly improved productivity (by 23% more volume) while also reducing maintenance and repair costs (by 26% decrease). Both of the aforementioned findings (similar or better performance and lower repair costs) further support the economics of simulator training. The lower cost of running simulator classes, with no fuel costs and the ability to accommodate more students per tutor, combined with the absence of costs associated with withdrawing machines from normal production, make this form of initial training more economical.

Further research should focus on compiling more detailed curricula for simulator training [17], improving technical solutions to alleviate motion sickness and emulate vibrations, and developing methods to assist training centers or employers in selecting operators with the best personal traits to create a productive and sustainable workforce [18,35]. It is important to note that the best operator is not necessarily the one who cuts the most trees. A good operator not only focuses on productivity but also takes good care of the machine’s maintenance, plans the work appropriately, and communicates effectively with colleagues. These skills require more time to develop than the short operator course.

5. Conclusions

• Simulator training is an efficient method for learning to operate forestry machines. The results of the study showed improvement in several aspects, including productivity, crane movement economy and control, and simultaneous boom movements.
• Most skill development occurred at the beginning of training. However, individual learning curves indicate that in many cases, learning continued throughout the course.
• Only the initial learning curves were analyzed. Complete personal development to become an experienced operator requires further training on a real machine.