Subtask-Based Usability Evaluation of Control Interfaces for Teleoperated Excavation Tasks

Abstract

This study aims to experimentally determine the most suitable control interface for different subtasks in the teleoperation of construction robots in a simulation environment. We compare a conventional lever-based rate control interface (“Rate-lever”) with two alternative methods: rate control (“Rate-3D”) and position control (“Position-3D”), both using a 3D positional input device. In the experiments, participants operated a construction machine in a virtual environment and evaluated the control interfaces across three tasks: sagittal plane excavation, turning, and continuous operation. The results revealed that “Position-3D” outperformed others for sagittal excavation, while both “Rate-lever” and “Rate-3D” were more effective for turning. Notably, “Position-3D” and “Rate-3D” can be implemented on the same input device and are easily integrated. This feature offers the possibility of a hybrid-type interface suitable for operators to obtain optimized performance in sagittal and horizontal tasks.

1. Introduction

1.1. Background

In recent years, the construction industry has been challenged by a shortage of skilled equipment operators and an aging workforce, leading to a decline in available labor. To address this and improve productivity, enhancing the operational efficiency of construction equipment has become essential [1,2,3,4]. Excavators, in particular, are indispensable at construction sites, as they are used not only for excavating earth and digging trenches but also for lifting heavy materials, grading land, and transporting loads. Given that excavator operations often occur in hazardous environments, ensuring operator safety is critical.

The integration of automation and teleoperation technologies in construction equipment presents opportunities to boost both productivity and safety. While automation reduces the need for manual labor, construction tasks often occur in dynamic, unpredictable environments, making full automation unfeasible. Human operators are still necessary for many tasks. Thus, a hybrid model combining automation and teleoperation is expected to be a practical solution for modern construction sites.

Teleoperation provides several key benefits. First, it enhances operator safety by allowing them to remain at a distance from hazardous environments [5], which is a crucial advantage not only for civil engineering but also in disaster response and recovery [6], where risks are high. Second, it reduces travel costs by enabling remote equipment control, eliminating the need for operators to be physically present on site. As teleoperation technology continues to develop, it has the potential to facilitate remote work on a global scale, further removing the constraints of time and distance.

However, there are some operability issues associated with teleoperation. There is concern that teleoperation may reduce operability and increase workload compared to normal direct control due to the lack of information and other factors [7]. In addition, not only teleoperation but also normal on-board control, the excavator’s control interface, is controlled by two levers handled by the left and right hands, depending on the tilt of the levers. The excavator’s control interface determines the velocity of the boom, arm, bucket, and swing according to the tilt of two levers handled by the left and right hands, which is not an intuitive control interface and is expected to enhance operability through habituation. This expectation is supported by the results of Morosi et al. [8]. This study reported that not only novice but also expert users were dissatisfied with the existing lever-based maneuvering interface. Therefore, a new maneuvering interface with higher operability than the conventional lever-based velocity control, which is called “Rate-lever”, is expected to be constructed.

To evaluate the control interface, we focus on the specific characteristics of the tasks performed by construction machines. As illustrated in Figure 1, typical excavator operations, such as excavation, turning, loading, and repositioning, consist of multiple subtasks [9,10]. In this study, we hypothesize that the optimal control method may vary depending on the subtask. For instance, when positioning the bucket for excavation, the conventional lever-based velocity control cannot directly specify the bucket’s exact position, often resulting in deviation from the target. Therefore, it may not be the most suitable method for precise positioning. Conversely, when turning the bucket to a target position after excavation, the conventional method—which relies on a timed tilting action—may be more appropriate.

1.2. Related Works

Several studies have explored the teleoperation of construction machinery with the goal of developing control interfaces that offer enhanced operability [8,11,12,13,14,15,16]. For example, Kim et al. developed a system that allows operators to control all movements of a small excavator with one hand, using sensors attached to the arm and fingers to detect movement and adjust the excavator arm accordingly [11]. Morosi et al. found that both velocity-input and position-input methods using a 3D input device provided more stable control than traditional methods with position-input proving particularly effective for the operator [8]. Similarly, Winck et al. designed an interface that links the rotation of the excavator’s hydraulic joints to the movement of the operator’s controls, allowing for more intuitive operation [12]. These studies demonstrate the potential of replacing conventional lever-based controls with multi-degree-of-freedom input systems to improve operator performance [13]. However, a significant limitation of these studies is that they evaluate teleoperation performance across tasks as a whole without differentiating between the specific subtasks involved. As a result, the optimal control method for each subtask remains unclear, which limits the practical applicability of these systems.

While some research outside the construction field has addressed the issue of task-specific control by switching between position and velocity control [17,18,19,20,21], similar approaches have not been investigated for the teleoperation of construction machinery. For example, Han et al. compared a hybrid control method that switches between position and velocity control to a position-only input system in the teleoperation of a manual drive mechanism with multiple subtasks [18]. They found that the hybrid system outperformed position-only control, as it could adapt to the specific demands of each subtask. Mokogwu et al. developed an algorithm that dynamically switches between position and velocity control based on the distance between the remote device and its environment [19]. These approaches highlight the potential for hybrid control systems but have not yet been applied to the teleoperation of construction equipment, presenting an opportunity for further exploration.

1.3. Objective

The objective of this study is to identify the optimal control interface for specific subtasks involved in the operation of construction machinery. Previous research has generally evaluated tasks as a whole without considering that different subtasks (e.g., digging, turning) may require distinct control methods for optimal performance. This study addresses this gap by comparing multiple control interfaces for specific subtasks, assessing their effectiveness in terms of operability and workload. Then, as a step toward future developments, we will explore the feasibility of proposing a hybrid control system that adapts to diverse operational requirements by integrating the findings of this study.

The novelty of this study lies in its task-specific evaluation of control interfaces. Unlike previous studies that have treated tasks as a unified whole, this research examines how different control methods perform in distinct subtasks. By conducting experiments to compare the performance of various control interfaces across different tasks in a simulated environment, we aim to determine the most effective control method for each subtask.

2. Method

The traditional control interface for excavators used at construction sites is based on lever-operated velocity control. This system, however, is often criticized for its lack of intuitiveness and steep learning curve. To address these issues, we propose an alternative maneuvering interface that employs a 3D input device capable of both position and velocity control. Figure 2 illustrates the correspondence between the 3D input device and the excavator for position control and velocity control. In position control mode, the input displacements d from the 3D device are directly mapped to the output displacements D of the excavator, allowing for precise control of its components. Conversely, in velocity control mode, the input displacements d are used to adjust the rate of movement of the excavator, facilitating dynamic adjustments and swift operational responses.

2.1. Conventional Method: Rate-Lever

A excavator operates with four degrees of freedom: the boom, arm, bucket, and swing. These components are moved by hydraulic cylinders that extend and retract to control each degree of freedom. The hydraulic actuators are manually controlled through proportional or servo valves, which regulate the flow of hydraulic oil from a pump to the specific cylinders.

Operators control the hydraulic flow rate using either four single-axis levers or two dual-axis levers. Figure 3 illustrates the standard control scheme for an excavator. Typically, the right hand manages the boom and bucket, while the left hand operates the arm and swing.

The control complexity arises from the non-intuitive mapping between the levers and the degrees of freedom of the excavator. When an operator adjusts a lever, it changes the position of the valve’s spool, which in turn alters the hydraulic oil flow rate to the corresponding cylinder. This change in flow rate affects the velocity of the hydraulic cylinder’s movement, and thus, the movement of the excavator’s components.

The velocity of each link in the hydraulic system is proportional to the tilt angle of the lever. For instance, the relationship between the lever position and the velocity of the boom’s hydraulic cylinder can be described as follows:

$${\mathit{v}}_{\mathit{l}}={\mathit{K}}_{\mathit{l}}{\mathit{u}}_{\mathit{l}},$$

(1)

$${\mathit{K}}_{\mathit{l}}=\left[\begin{array}{cccc}{k}_{l,1}& 0& 0& 0\\ 0& k_{l,2}& 0& 0\\ 0& 0& k_{l,3}& 0\\ 0& 0& 0& k_{l,4}\end{array}\right],$$

(2)

where ${\mathit{v}}_{\mathit{l}}$, ${\mathit{K}}_{\mathit{l}}$, and ${\mathit{u}}_{\mathit{l}}$ are the link cylinder velocities, the lever angles, and the gains corresponding to the cylinders, respectively. The gains were determined empirically by the authors.

2.2. Comparative Method 1: Position-3D

This section details the position control of an excavator using a 3D input device. The study introduces a cooperative control approach that utilizes a 3D input device to enhance operability compared to traditional control methods.

We employed a position input interface (3D Systems, Phantom Premium) which features a force presentation function. This device is capable of providing 6-DOF (degrees of freedom) position and orientation input as well as delivering 3-DOF translational forces. For the excavator control, the interface facilitates precise manipulation of the boom and arm, which are assumed to operate in tandem, while the swing and bucket are controlled independently.

At first, we address the maneuvering of the boom and arm. Figure 4 illustrates the reference scheme used for kinematic analysis. For a multi-axis hydraulically driven arm, it is necessary to determine the target lengths of the hydraulic cylinders for the boom and arm based on the desired joint angles.

The target cylinder lengths for the boom and arm, denoted as $s_1$ and $s_2$, respectively, are calculated as follows:

$$s_1=\sqrt{{l_{12}}^2+{l_{13}}^2-2l_{11}{l}_{12}cos{\varphi }_1}-c_1,$$

(3)

$$s_2=\sqrt{{l_{22}}^2+{l_{23}}^2-2l_{21}{l}_{22}cos{\varphi }_2}-c_2.$$

(4)

The target cylinder length for each hydraulic actuator is determined and then controlled by applying a gain to the difference between this target length and the current measured cylinder length.

Next, we describe the swing operation. In this process, the angle between the position of the 3D input device and the origin in the planar coordinate system of the virtual environment serves as the target value. The angle between the current position of the arm tip and the origin is the measured value. The control of the swing motion is achieved by calculating the difference between the target angle and the measured angle and using this difference to adjust the swing movement.

Finally, we address the bucket operation. The rotation angle of the bucket is controlled by the rotation angle of the handle on the 3D input device. The hydraulic cylinder is configured to be at its maximum length when the 3D input device handle reaches its maximum rotation angle and at its minimum length when the handle is at the zero position. The relationship between the bucket angle and the cylinder length is approximated linearly. The target value for the bucket cylinder length is set according to the rotation angle of the 3D input device, and the current bucket cylinder length is the measured value. The control of the bucket cylinder is then adjusted based on the difference between the target and measured values. Figure 5 illustrates the correspondence between the 3D input device and the excavator in the Position-3D control scheme. Position mapping is performed with reference to the respective local coordinate system.

The equation for the velocity of the hydraulic cylinder controlling the boom in the Position-3D maneuvering scheme is given below.

$$\mathit{x}=k_p\mathit{u},$$

(5)

where $\mathit{x}$, $k_p$, and $\mathit{u}$ are the 3D bucket target positions, the gain, and the 3D input positions, respectively. The gain was determined empirically. The rotation angle of the bucket is proportional to the angle of the rotating part of the input device gripper.

2.3. Comparative Method 2: Rate-3D

In the Rate-3D maneuvering scheme using a 3D positional input device, the boom, arm, and swing operations are divided into inactive and active zones, as illustrated in Figure 6.

When the tip of the 3D input device moves outside the inactive zone, the vector between the tip of the device and the center of the inactive zone is read. The excavator’s arm tip then moves in the direction of this vector. The excavator will continue to move in that direction as long as the tip of the device remains within the active zone, and it will cease movement when it returns to the inactive zone.

The control algorithm first updates the current position coordinates by adding the read vector to them, setting this as the new target position. From this target position, the required cylinder lengths for the boom and arm are calculated similarly to the Position-3D method.

For swing control, the vector representing the lateral direction of the 3D input device is read, and the magnitude of this vector determines the input.

Bucket maneuvering also involves an inactive range. Beyond this range, the bucket moves with a constant input.

The equation for the velocity of the hydraulic cylinder controlling the boom in the Rate-3D maneuvering scheme is given below:

$$\mathit{v}=k_r\mathit{u},$$

(6)

where $\mathit{v}=(v_x,v_y,v_r)$, $k_r$ and $\mathit{u}$ are the target velocities, the gain, and the 3D input positions, respectively. $v_x$ and $v_y$ are two-dimensional information, the sagittal plane bucket positions, and $v_r$ is the rotation angle of the robot body. The gain was determined empirically. The rotation angle of the bucket is proportional to the angle of the rotating part of the input device gripper.

To help the operator easily determine whether the position of the 3D input device is within the inactive or active areas, a reaction force is applied. This reaction force returns the device to the center of the inactive area when it moves into the active area, as illustrated in Figure 6.

The reaction force is defined by the following equations:

$$\mathit{F}=k_f\mathit{u},$$

(7)

where $\mathit{F}$ and $k_f$ are the reaction force presented to the operator via the 3D input device and the gain. The gain of the reaction force presentation was determined empirically so that the operator perceives the reaction force around the time when he/she leaves the immobile region.

3. Experiment

3.1. Condition

The objective of this experiment was to evaluate and compare the performance of three maneuvering methods for teleoperating an excavator, focusing on their effectiveness for different types of work. The methods compared were “Rate-lever”, “Rate-3D”, and “Position-3D”. This comparison aimed to assess how each method performs in various operational tasks.

Figure 7a shows the experimental condition using the conventional rate-lever method. The input devices are independent left and right lever-type controllers (T.16000M FCS, Thrustmaster, Hillsboro, OR, USA). In contrast, the methods utilizing the 3D input device are illustrated in Figure 7b. The input device is a 3D-pointing haptic device (Phantom Premium, 3D Systems, Rock Hill, SC, USA) used with one hand. During the experiment, participants operate the excavator while viewing two monitors: one from the overhead viewpoint and the other from the operator’s viewpoint.

The study involved a total of 10 participants, comprising 8 males and 2 females, all in their 20s. Of the participants, 9 were right-handed and 1 was left-handed. None of the participants had prior experience operating construction machinery.

3.2. Tasks

Since excavator tasks can be performed in the sagittal plane, involve turning, or combine both actions, three specific tasks were selected for this experiment.

3.2.1. Task 1: Sagittal Plane Work

The first task involves sagittal work that simulates slope shaping. Figure 8 illustrates Task 1.

In this task, the operator moves the bucket tip along a designated target trajectory. The sequence of actions is as follows:

1.

The operator first moves the bucket to target object 1. As the bucket approaches target object 1, its color changes from yellow to red.

2.

Next, the operator moves the bucket from target object 1 to target object 2.

3.

The bucket is then moved from target object 2 to target object 3.

4.

Finally, when the bucket tip passes over target object 3, all target objects on the screen change from red back to yellow, signaling the completion of the task.

3.2.2. Task 2: Turning Operation

The second task involves a turning operation, requiring a 180° turn. Figure 9 illustrates Task 2.

In this task, the operator is required to turn the excavator from a specified starting position to another designated location. The steps are as follows:

1.

The excavator is initially swung diagonally to the left from its starting position to Waypoint 1, which is located in front of and to the left of the excavator.

2.

When the bucket makes contact with Waypoint 1, its color changes from blue to red.

3.

The excavator then swings the bucket to Waypoint 2. Upon contact with Waypoint 2, its color changes from blue to red, indicating the completion of the task.

3.2.3. Task 3: Series of Multiple Subtasks

The third task is an excavation operation, involving excavation within a designated area, and encompasses several subtasks performed in sequence. Figure 10 illustrates Task 3.

In this task, the following steps are performed:

1.

One of three excavation areas is randomly selected and displayed.

2.

The operator moves the bucket to a randomly presented area. As the bucket approaches Waypoint 1, its color changes from green to red.

3.

The bucket is then moved to the excavation area, where excavation is carried out.

4.

After excavation, the bucket is lifted. When the bucket reaches a predetermined height, all waypoints change from green to red, indicating the completion of the task.

3.3. Procedure

The experimental procedure was conducted as follows:

1.

At first, each of the three types of maneuvering methods was explained as a preliminary step to the experiment.

2.

The content of the tasks was then explained to the participants, and the testing proceeded in the order of Task 1, Task 2, and Task 3.

3.

To familiarize the participants with the maneuvers, a 10-min practice period was provided before each test.

4.

For each task, the order for each maneuvering method was randomized with three trials performed for each maneuver.

5.

After completing three trials of each task, participants were asked to fill out a questionnaire.

3.4. Evaluation Items

Two types of evaluations were conducted: objective and subjective.

3.4.1. Objective Evaluation

The objective evaluation included the following metrics:

• Working time: The time required to move from each target object to the next in Task 1.
• Trajectory error: The root-mean-square error between the target trajectory and the actual trajectory in Task 1.
• Excavation volume: The amount of soil excavated by the excavator in Task 3.

3.4.2. Subjective Evaluation

The subjective evaluation, detailed in

Table 1. Questionnaire of experiment.

Item	Adjectives (1/5)	Text
Operability	Low/high	How easy was it to operate?
Ease of learning	Low/high	How easy was it to learn?
Physical demand	Low/high	How physically demanding was the task?
Mental demand	Low/high	How mentally demanding was the task?

, consisted of four items: operability, ease of learning, physical demand, and mental demand. Participants rated each item on a 5-point scale. These questionnaires were determined with reference to NASA-TLX [22].

3.5. Results

3.5.1. Task 1: Sagittal Plane Work

Working Time

Figure 11 shows the working time for the three maneuvering methods.

A Friedman test was performed to compare the working times among the three methods with a significance level set at $\mathit{p}<0.05$. The results indicated a significant difference among the three control methods.

To identify which specific methods differed significantly, a Wilcoxon signed rank sum test was conducted with a significance level of $\mathit{p}<0.05$, and Bonferroni’s correction was applied. The results showed the significant differences between “Rete-lever” and “Position-3D” ($\mathit{p}<0.05$) in route from 2 to 3. In addition, significant differences were found between “Rate-lever and “Rate-3D” ($\mathit{p}<0.01$) and “Rate-lever” and “Position-3D” ($\mathit{p}<0.01$) in route from 1 to 3. The findings suggest that Position-3D generally requires less time when the operation involves following a target trajectory, such as in sagittal plane work from 2 to 3.

Trajectory Error

Figure 12 shows the orbit errors between the actual and target orbits for the three maneuvering methods.

A Friedman test was conducted to compare the trajectory errors among the three maneuvering methods. The results indicated that there was no significant difference in the trajectory error ($\mathit{p}>0.05$). However, the mean value of orbit error tended to be smaller for Position-3D compared to the other maneuvering methods.

Operability

Figure 13 shows the operability evaluation for the three maneuvering methods.

A Friedman test was performed to assess the differences in operability among the three methods. The results indicated a significant difference among the control methods ($\mathit{p}<0.05$). To further investigate which methods differed significantly, a Wilcoxon signed rank sum test was conducted, and Bonferroni’s correction was applied. Significant differences were found between “Rate-lever” and “Rate-3D” ($\mathit{p}<0.05$), “Rate-lever” and “Position-3D” ($\mathit{p}<0.01$), and “Rate-3D” and “Position-3D” ($\mathit{p}<0.01$). These findings suggest that “Position-3D” is perceived as more intuitive compared to the other maneuvering methods for tasks involving sagittal plane work.

Ease of Learning

Figure 14 shows an evaluation of the ease of learning the three maneuvering methods.

A Friedman test was performed to assess the differences in proficiency among the three methods. The results indicated a significant difference among the control methods ($\mathit{p}<0.05$). To further identify which methods differed significantly, a Wilcoxon signed rank sum test was conducted, and Bonferroni’s correction was applied. Significant differences were found between “Rate-lever” and “Rate-3D” ($\mathit{p}<0.05$), “Rate-lever” and “Position-3D” ($\mathit{p}<0.01$), and “Rate-3D” and “Position-3D” ($\mathit{p}<0.05$). These findings suggest that the proposed “Position-3D” method is more proficient than the conventional “Rate-lever” method for sagittal plane work. Additionally, “Position-3D” demonstrates superior proficiency compared to “Rate-3D”.

Physical Demand

Figure 15 shows the evaluation of physical demand for the three maneuvering methods.

A Friedman test was performed to assess the differences in physical exhaustion among the three methods. The results indicated that there was no significant difference among the three maneuvering methods ($\mathit{p}>0.05$). However, the mean physical fatigue tended to be higher for the maneuvers using the 3D input device compared to those using the lever in sagittal plane work.

Mental Demand

Figure 16 shows the evaluation of mental demand for the three maneuvering methods.

A Friedman test was conducted to assess the differences in mental demand among the three methods with the significance level set at $\mathit{p}<0.05$. The results indicated that there was a significant difference among the three maneuvering methods ($\mathit{p}<0.05$).

Subsequently, a Wilcoxon signed rank sum test was performed to identify which methods significantly differed from each other, and Bonferroni’s correction was applied. The results revealed significant differences between “Rate-lever” and “Position-3D” ($\mathit{p}<0.05$) and “Rate-3D” and “Position-3D” ($\mathit{p}<0.05$). This indicates that “Position-3D” results in the lowest mental demand during sagittal plane work.

3.5.2. Task 2: Turning Operation

Operability

Figure 17 shows the operability evaluation for the three control methods.

A Friedman test was conducted to assess operability differences among the three methods. The results revealed significant differences among the three control methods ($\mathit{p}<0.05$).

Subsequently, a Wilcoxon signed rank sum test was performed to determine which methods had significant differences, and Bonferroni’s correction was applied. The results showed significant differences between “Rate-lever” and “Position-3D” ($\mathit{p}<0.01$) and “Rate-3D” and “Position-3D” ($\mathit{p}<0.01$). This indicates that “Rate-lever” and “Rate-3D” are more intuitive than “Position-3D” for turning tasks.

Ease of Learning

Figure 18 shows an evaluation of the ease of learning the three maneuvering methods.

A Friedman test was conducted to assess the ease of learning differences among the three methods. The results indicated that there was no significant difference among the three maneuvering methods ($\mathit{p}>0.05$). This suggests that the differences in the ease of learning for turning operations were minimal across the three control methods.

Physical Demand

Figure 19 shows the evaluation of physical demand for the three maneuvering methods.

A Friedman test was performed to assess physical demand across the three maneuvering methods. The results revealed a significant difference among the three methods ($\mathit{p}<0.05$).

To identify which methods differed significantly, a Wilcoxon signed rank sum test was conducted with Bonferroni’s correction applied. The results showed significant differences between “Rate-lever” and “Position-3D” ($\mathit{p}<0.01$) and between “Rate-3D” and “Position-3D” ($\mathit{p}<0.01$). This indicates that physical fatigue was higher for “Position-3D” compared to the other two methods during the turning task.

Mental Demand

Figure 20 shows the evaluation of mental demand for the three types of maneuvers.

A Friedman test was conducted to assess mental demand across the three maneuvering methods. The results revealed a significant difference among the three methods ($\mathit{p}<0.05$).

To determine which methods had significant differences, a Wilcoxon signed rank sum test was performed with Bonferroni’s correction applied. The results showed significant differences between “Rate-lever” and “Position-3D” ($\mathit{p}<0.05$) and between “Rate-3D” and “Position-3D” ($\mathit{p}<0.05$). This indicates that both “Rate-lever” and “Rate-3D” were less mentally demand than “Position-3D” during the turning task.

3.5.3. Task 3: Series of Multiple Subtasks

Excavation Volume

Figure 21 shows the excavation volume for the three maneuvering methods.

A Friedman test was performed to compare the excavation volumes among the three maneuvering methods. The results indicated a significant difference among the methods ($\mathit{p}<0.05$). To determine which methods differed significantly, a Wilcoxon signed rank sum test was conducted with Bonferroni’s correction applied. The results revealed a significant difference between “Rate-lever” and “Position-3D” ($\mathit{p}<0.05$). This suggests that the excavation volume was greater for “Position-3D” compared to “Rate-lever” during excavation operations.

Operability

Figure 22 shows the operability evaluation for the three maneuvering methods.

A Friedman test was conducted to assess operability across the three methods. The results indicated a significant difference among the methods ($\mathit{p}<0.05$).

To identify which methods differed significantly, a Wilcoxon signed rank sum test was performed with Bonferroni’s correction applied. The results revealed significant differences between “Rate-lever” and “Position-3D” ($\mathit{p}<0.05$) and “Rate-3D” and “Position-3D” ($\mathit{p}<0.05$). This suggests that “Position-3D” is more intuitive than the other two methods for excavation tasks.

Ease of Learning

Figure 23 shows the evaluation of ease of learning the three maneuvering methods.

A Friedman test indicated a significant difference among the methods ($\mathit{p}<0.05$).

To identify which methods differed significantly, a Wilcoxon signed rank sum test was conducted with Bonferroni’s correction applied. The results showed a significant difference between “Rate-lever” and “Position-3D” ($\mathit{p}<0.05$). This indicates that “Position-3D” is more proficient than “Rate-lever” for excavation tasks.

Physical Demand

Figure 24 shows the evaluation of physical demand for the three maneuvering methods.

A Friedman test was conducted to assess physical demand among the three maneuvering methods. The results indicated a significant difference among the methods ($\mathit{p}<0.05$).

To determine which methods differed significantly, a Wilcoxon signed rank sum test was performed with Bonferroni’s correction applied. The results showed significant differences between “Rate-lever” and “Rate-3D” ($\mathit{p}<0.05$) and “Rate-lever” and “Position-3D” ($\mathit{p}<0.05$). This indicates that physical demand is higher for “Rate-3D” and “Position-3D” compared to “Rate-lever” in excavation total operations.

Mental Demand

Figure 25 shows the evaluation of mental demand for the three control methods.

A Friedman test was conducted to assess mental fatigue among the three maneuvering methods. The results indicated that there was no significant difference among the three methods $\mathit{p}>0.05$. This suggests that the differences in mental fatigue among the three control methods were minimal during excavation operations.

4. Discussion

4.1. Integration of Suitable Control Methods for Each Subtask

Based on the results obtained from the experiments, the optimal control interface for each task was discussed below.

In Task 1, the objective evaluation indicated that “Position-3D” tended to reduce both the work time and trajectory error compared to other maneuvering methods, although the differences were not statistically significant. This was corroborated by the subjective evaluation, which also favored “Position-3D”. Consequently, “Position-3D” with a 3D input device is considered more suitable for sagittal plane work. In Task 2, subjective evaluations revealed that “Rate-lever” and “Rate-3D” outperformed “Position-3D”. The “Position-3D” method requires extensive arm movements when performing turns, which complicates maneuvering. In contrast, “Rate-lever” and “Rate-3D” involve minimal arm movement, making them more straightforward for turning tasks. Thus, “Rate-lever” and “Rate-3D” are deemed more effective for turning operations.

In summary, “Position-3D” is superior for sagittal work, while “Rate-lever” and “Rate-3D” are more effective for turning tasks. Given that “Position-3D” and “Rate-3D” can be used with the same device, an integrated interface was developed as shown in Figure 26. This integration provides a single maneuvering interface with high operability, which is capable of switching between control methods for different tasks.

4.2. Physical Demand Reduction

The experimental results indicate that the control method using the 3D positional input device results in higher physical demand compared to the conventional lever-based method. This increased fatigue is attributed to the continuous raising of the arms when using the 3D input device.

To address this issue, several countermeasures can be considered to reduce physical fatigue. For example, installing armrests can alleviate the strain on the arms. Klauer et al. [23] and Jackson et al. [24] utilized robotic arms as armrests to support patients’ movements during rehabilitation. Additionally, Ott et al. [25] demonstrated that gravity compensation can reduce muscle fatigue when using a 3D input device.

Given these findings, the installation of armrests to reduce operator arm fatigue will be explored in future studies.

4.3. Integration with Haptic Feedback and Guidance

An effective method to enhance the operability of teleoperation is to integrate haptic feedback and guidance approaches.

Haptic feedback is a method of presenting haptic information generated by contact with the environment to the operator to aid understanding of the environment and improve workability. Force feedback is the most typical method [26,27]. For instance, the reaction force generated when a bucket makes contact with the environment is measured or estimated [28], and then it is reproduced by a device that can present the force sensation. The manipulation device used in this research is capable of presenting force, and integration with such force feedback is easy. If force feedback were to be realized, the unstability of the system and the effects of handshaking might appear, and this should be a concern.

Another type of haptic feedback is vibrotactile feedback [29,30,31], which utilizes high-frequency vibration information associated with contact. This method uses an accelerometer to measure the vibration near the bucket and reproduces it on a vibrotactile display. This method is expected to be used in combination with force feedback because it includes information that differs from force feedback, such as differences in the material of the sand and soil.

Haptic guidance lead the robot toward a target position or trajectory. For instance, the system can provide force feedback proportional to the error between the target position, as determined by the system, and the current input position, assisting in guiding the operator to the target position. Such assistance is expected to improve the operator’s performance compared to normal operation.

In a haptic guidance approach, controlling construction equipment along a trajectory similar to that of a skilled operator is desirable [32]. Lee et al. [33] developed an algorithm that tracks the bucket tip to a target position, demonstrating reduced error compared to positional control alone in ground leveling operations. Additionally, Son [34] showed that the digging performance can be optimized by generating a trajectory that mimics expert behavior.

5. Conclusions

In this study, multiple control methods were compared based on criteria representing operability and workload to determine the optimal control method for each subtask, such as digging and turning. First, we implemented three types of control methods in a virtual environment for comparison. The three methods were the conventional method of velocity control using a lever (Rate-lever), velocity control using a 3D positional input device (Rate-3D), and position control using a 3D input device (Position-3D). Next, experiments were conducted to compare these control methods for each task and evaluate their performance. The results demonstrated that “Position-3D” was superior for sagittal plane work. This is probably because sagittal plane work requires positional accuracy, and the motion range does not need to be that wide. The result also indicated thar “Rate-lever” and “Rate-3D” were shown to be superior in turning tasks. This is presumably due to the characteristics whereby turning work moves over a wide motion range and does not require positional accuracy. Based on these findings, the possibility of a hybrid control method in which a single input device switches the control method depending on the task, such as “Position-3D” control for sagittal work and “Rate-3D” control for turning work, was discussed.