Keywords

1 Introduction

Ergonomic simulation enables a lot of possibilities to improve the ergonomics and productivity of manual production processes. Despite its potential, ergonomic simulation is used quite rarely in production planning processes compared to other simulation methods e.g. robotic simulation. The main reason for this is the great amount of time required to control a digital human, which, in turn, is often needed to create human movements in the simulation [1]. Basically, it is possible to set a posture of a digital human model by forward or by inverse kinematics. Using forward kinematics, the user has to set every single joint; this requires experience and time. On the other hand, inverse kinematics based on posturing is a fast method. It, however, frequently leads to unsatisfactory postures. Apart from these methods, it is possible to use motion capture systems for controlling the digital human, often requiring preparation time and effort as well as physical prototypes of product and equipment.

To reduce the time needed for posturing in human simulation, we have been following a new approach. Consequently, we have developed a special input device. A small physical puppet is used to control the posture of a digital human model. We already introduced the prototype of our input device, named Human Input Device (HID) elsewhere [2]. In this paper, we want to present our sophisticated version and the evaluation of our input device.

2 Related Work

There have been several other projects where graspable input devices controlled the posture of virtual rigged characters [3,4,5,6,7,8,9,10,11,12,13,14]. Most of them focus on computer animation e.g. the Dinosaur Input Device [3], which was one of the first entities of its kind in input devices. It was made to control virtual dinosaurs playing in the movie Jurassic Park. Monkey [4] was the first humanlike input device for controlling a digital human. It was used for ergonomic simulation in product design and controlled a very rudimentary digital human. A more modern input device for ergonomic simulation in product design was presented by Yoshikazi et al. [5]. Their actuated puppet also controlled a very simple digital human with only a few degrees of freedom compared to modern human models used in ergonomic simulation.

Moreover, an evaluation has not yet been made whether if it is useful to control a digital human for ergonomic simulation of production processes with such a graspable user interface. Yoshikazi et al. [5] conducted a study to evaluate their input device for posturing processes in general. But they used their input device to control a simple virtual artist’s doll. Today’s digital human models used in ergonomic simulation are more complex. Furthermore, the posturing process did not represent manual production processes.

3 Design

The dimensions of the puppet correspond with the 50th percentile of the German population and is scaled by the factor 1:4.25. The puppet is an input device with 22 degrees of freedom and is designed to control the digital human model Jack by Siemens PLM. Thus, we have developed a plugin to use in the simulation tool process simulate. In comparison to related input devices, our input device is able to control a posture of a digital human model with 77 degrees of freedom. Hence, we have now designed a suitable skeleton for the puppet derived from the Jack model. As a result the postures of Jack and the puppet nearly match, although the input device has only 22 degrees of freedom (see Fig. 1).

Fig. 1.
figure 1

Setting a posture with the human input device to the digital human model Jack

Full size image

3.1 Mechanics

Using the Jack model, we determined the skeleton of the puppet. Figure 2 shows which joints can be manipulated, where a schematic (left) and the final version (right) of the skeleton is shown.

Fig. 2.
figure 2

Schematic (left) and final (right) version of the skeleton

Full size image

We tried to separate the mechanical structure into functional parts and shaping parts. For the functional parts, we designed a special bearing, which encapsulates an encoder and represents the center of each joint of the puppet (see Fig. 3). The bearing is designed for conventional manufacturing processes and it is equal for every joint. With two adjusting screws, it is possible to set the friction of the joint for compensating gravity.

Fig. 3.
figure 3

Joint structure

Full size image

The individual shaping parts are connected to the bearings and ensure that the puppet looks as similar as possible to Jack. These parts are designed for modern additive manufacturing processes. Due to the encoder, it is only possible to measure a single degree of freedom per each joint. Hence, we had to substitute three dimensional joints e.g. the hip by three serial connected joints. In addition, we simplified the spine and the clavicle-shoulder group by three serial connected joints each. The values of the joints are used for controlling the whole group including e.g. each vertebra of the spine. Consequently, the posture of the digital human model still looks natural. Figure 4 (left) shows the shape of HID compared to Jack.

Fig. 4.
figure 4

Shape of the HID compared to Jack (left) and customized board (right)

Full size image

3.2 Interface

We developed a small customized board, including a microcontroller to decode the signals of the encoder (see Fig. 4, right). These boards are integrated inside the shaping parts of the puppet. Each Microcontroller monitors the signals of one encoder and converts them to an angle. Furthermore, they are slave participants to an I2C bus.

We subsequently developed a controller box, which contains a main board. Through an additional microcontroller on it, representing the master participant to the bus, the joint values are provided to our plugin in process simulate. Finally, our plugin, based on the API of process simulate, maps the posture of the puppet to Jack with no noticeable latency. Furthermore, the user can use the plugin for saving characteristic poses to generate keyframe based movements of the digital human model.

4 Evaluation

We conducted a study to see if our input device is suitable for ergonomic simulation of production processes. The study is designed to see if our input device has higher usability along the learning curve for controlling the posture of a digital human model for ergonomic simulation compared to the mouse. The following research questions thus needed to be answered:

  • RQ1: Are users along the learning curve more effective if they use the HID instead of the mouse for posturing a digital human model.

  • RQ2: Are users along the learning curve more efficient if they use the HID instead of the mouse for posturing a digital human model.

  • RQ3: Are users along the learning curve more satisfied if they use the HID instead of the mouse for posturing a digital human model.

4.1 Design of the Study

Task

In our study users should perform a core task in ergonomic simulation. They should manipulate the posture of the digital human model Jack to different target postures. We defined in total six different target postures, which all considered manual production processes (see Fig. 5). Thus, the digital human model interacts with equipment and products. One half of the postures represents final postures of a movement e.g. reaching a part in a container and the other half represents intermediate postures of movements. One part was defined with the HID and the other part with the mouse. The six scenes consisting of digital human model, equipment and product are duplicated and ordered next to the original scene. But the posture of Jack is set to a neutral default posture. The users now had to manipulate the default postures to the target postures repeatedly with mouse and HID. The users should terminate the posturing process if they are satisfied with the approximation of the manipulated and target posture or, if they notice that they are not making any progress in accuracy. Manipulating the posture with the mouse, users used the normal graphical user interface in process simulate. They could decide whether to use direct or inverse kinematics. Taking a learning curve into consideration, each participant repeated the run seven times. The first six runs with a maximal distance of two days and the last run with a distance of two weeks to the sixth run to consider the reusability.

Fig. 5.
figure 5

Target postures

Full size image

Participants

We conducted the study with 14 participants. Eleven of them were male and three female. They were between 18 and 34 years old. All of them were students of mechanical engineering and had no experience in ergonomic simulation.

Order Balancing

To minimize the effect on the posturing process of which input device is used first for manipulating the posture, we divided the participants into two different groups. We consequently altered the type of input device which was used first. The order of group A is the inversion of group B (see Fig. 6).

Fig. 6.
figure 6

Alter of the first input device for each posturing process of both groups

Full size image

Variables

The independent variables, which we changed in this study, are the type of input device, the different runs and the participants. The dependent variables are the three components of usability: effectiveness, efficiency and satisfaction. The controlled variables are the target postures and the order in which an input device is used for posturing.

Operationalization of the Dependent Variables

For measuring the dependent variables, we had to derive values which represent them and which can be measured. To measure the accuracy, we determined each deviation of the target posture and the manipulated pose. Thus, we summed up the absolute value of the difference for every joint between the target posture and the manipulated posture. With each run we noted the average accuracy out of all six postures made with the mouse and HID. We also measured the execution time of the posturing processes to consider the efficiency. For every run, we determined the average execution time out of all six postures achieved with the mouse and HID. We then asked the participants after each run how satisfied they were while using the two types of input devices. They had to rate their satisfaction on a scale from one to ten, where ten being high and one is low.

Hypotheses

For each run we formulated the following three hypotheses:

RQ1:

  • H0: The achieved accuracy with the mouse is higher or the same

  • H1: The achieved accuracy with the mouse is less

RQ2:

  • H0: The achieved execution time with the mouse is shorter or the same

  • H1: The achieved execution time with the mouse is longer

RQ2:

  • H0: The achieved user satisfaction with the mouse is higher or the same

  • H1: The achieved user satisfaction with the mouse is lower

4.2 Procedure of the Study

Before the first run, the users where introduced to both input devices, which took ca. 30 min. First they were shown how to use the input devices and then they could try them out. The procedure was carried in a normal office at a work station equipped with two monitors. On the first monitor they had a 3d-view and on the second the graphical dialogs of the input devices. The HID were initially positioned next to the monitor, but the user were given explicit permission to relocate them as they wished. Moreover, they could choose whether to stand or sit while using the HID.

4.3 Results

Figures 7, 8 and 9 show the average values and confidence intervals (95%) of the average execution time, accuracy and the rating of the user satisfaction per each run and input device.

Fig. 7.
figure 7

Average values and confidence intervals (95%) of the average task execution times per run

Full size image
Fig. 8.
figure 8

Average values and confidence intervals (95%) of the average accuracy per run

Full size image
Fig. 9.
figure 9

Average values and confidence intervals (95%) of the user satisfactions per run

Full size image

A Shapiro-Wilk-Test showed that we can assume that one part of the variables is distributed normally and another part is not. We verified the hypotheses with paired sample t-tests or Wilcoxon signed-rank tests depending on whether normally distributed variables are considered or not. Table 1 shows the results of the multiple testing. We used the Bonferroni-Correction to compensate the increase in type 1 error, due to multiple testing. Hence, we divided our total α = 0.05 by the number of hypotheses (21), which leads to a partial type 1 error αp = 0.00239 for each hypothesis.

Table 1. Test results
Full size table

5 Conclusion and Future Work

The test results show that the participants where significantly faster in posturing when using the HID instead of the mouse. This result was shown along the whole monitored learning curve. Although all average values of accuracy or user satisfaction achieved with the HID where higher than the average values achieved with the mouse, it could not be shown that these values where significantly higher. Nevertheless, what was clear was that using the HID minimizes the effort required for posturing a digital human model in ergonomic simulation for inexperienced users in the learning process. To verify that it is useful to use the HID for ergonomic simulation, we are now planning to conduct an additional study with domain experts.