How Do We Sit When Our Car Drives for Us?

Digital human models (DHMs) help to include necessary ergonomic and safety aspects early in the product development process. With the introduction of automated driving in SAE Levels 3 and 4 new challenges in automotive design arise and established DHMs need to adapt to new postures and tasks [1]. The driver becomes a passenger, thus new aspects in comfort [2] and safety [3] are researched. The posture of the participant in those study is often estimated after relevant tasks. The missing link between the task and the according posture in SAE level 3 and 4 automated driving [4] is evaluated in this study.

Since it is hard to conduct experiments regarding the passenger in real cars, studies were conducted in trains to gather first insights. Body postures were recorded through notes of the experimenter and later put into larger categories. The studies showed a huge variety in body postures [5, 6]. Since the new freedom of the passenger in a level 3 automation can be interrupted by a take-over request, [7] found that the take-over quality does not suffer under a more open knee joint angle of 133° (compared to 114° for conventional vehicles). A flatter backrest angle (38° compared to 24° in conventional vehicles) leads to a worse performance in the hands-on-time.

The study was approved by the ethics committee of the Technical University of Munich. The experiment was conducted at the Modular Ergonomic Mock-Up [8] at the Chair of Ergonomics of the Technical University of Munich, which is an automated multifunctional version of a seat mock-up and is based on a structure of parallel rail elements. The setup consisted of a car seat, pedals and the steering wheel. Two configuration were tested, the “Level 3” and the “Level 4” condition. Both are meant to represent a state according to SAE automated driving levels. For “Level 3” seat, pedals and steering wheel were configured in an SUV setting with an average H30 of 350 mm. The Backrest angle was fixed to 22°. It was prohibited to touch the steering wheel with either hand or objects during tasks. The participant was instructed to adjust the seat into a driving position, then the seat was moved backwards until the knee had an opening angle of 133° (see Fig. 1). This posture should allow an adequate take-over-performance as it is needed in a level 3 automated driving [7], while giving the passenger a little more space. For the condition “Level 4” the pedals and steering wheel were removed, thus representing a level 4 automated driving, where this could be feasible car environment.

Experimental setup

Full size image

For those two conditions every participant was asked to perform the eleven tasks in Table 1 each for two minutes in a randomized order. The task “working with laptop” could not be conducted, because of the lack of space in the mock-up and was correspondingly dropped from the “Level 3” condition.

The postures chosen by the participants were filmed. The experimenter later classified the material with seven digits for four body regions; the head, the torso, the hands, the legs (see Fig. 2). The possible outcome for the digits are described in Tables 2, 3, 4 and 5 and were iteratively developed during the coding thus leaving little data unusable. Since some participants changed their posture during a task, it is possible that some participants contribute more postures to the dataset than others. A posture was classified after 4 s of holding, thus eliminating smaller tasks like fixing hair or scratching the nose.

Example coding for one posture

Full size image

25 students and research associates owning a German driving license between the age of 21 to 30 years (M = 25.2; SD = 2.33) participated in the experiment for pay. The sample consists of 11 females and 14 males. The body height range extends from 1585 mm to 1900 mm.

Table 2 shows the classification for the body part head, containing head and headrest. For both there is the “difficult to detect” option, which was introduced after some tasks resulted in a constant moving of the head. “Straight” defines a head position where the participant is looking straight forward. As soon as the head was rotated sideways it was classified so ignoring the aspects of vertical orientation.

Table 3 shows the classification for the torso. Similar to the Head one digit describes the torso itself the other relation of the torso to the backrest. A slightly reclined posture of the upper body against the backrest is defined as the start (neutral) position of the torso. So the neutral posture corresponds to a conventional vehicle design today, where torso angles of 22° to 25° are used. In reality those angles are observed to range from 5° to 35° [9]. Forwards and backwards an in relation to the neutral posture. A backwards leaning torso results from moving the hip forward out of the seat.

Table 4 shows the classification of the hands. Left- and right-handedness was ignored, by coding the hand dominating the task with “Hand A”. For tasks, where an object is held in the hand, this hand was considered dominant. While the number gives the height of the hand, the letter gives information about the activity of the hand.

During the coding, a priority order is ensured, which means that a working hand is numbered prior to a resting hand. Also, a resting hand has priority over a hand supporting a body part, such as the example of “4w2s”. In the case, that the two hands cannot be distinguished through dominance or status, the one with the lower position is then placed first, like the example of “2s4s” shows. This being a combination with repetition creates 91 possible combinations of hand postures.

Table 5 shows the classification for the legs, which differentiates mainly between the flexed/extended states. Knee joint angles seen from the side and below 90° are regarded as flexed, while above 110° was classified with extended.

Using this coding method a total of 1099 body postures were identified, 497 for the “Level 3” and 602 for the “Level 4” condition. 232 full body combinations were found for “Level 3” and 299 for “Level 4”. Specifically less than 5 times (n < 5) occur 210 combinations in the “Level 3” condition and 278 in the “Level 4” condition. This is comparable to white noise in the data when looking at the whole body. Thus the data is analyzed for the four body regions separately for the two conditions in Sect. 3.1. Sect. 6 displays the results for every body region according to task and condition, which are explained in Sect. 3.2.

3.1 Condition Specific Results

Head

For the head the conditions were ignored as they had no visible effect. As per Table 6 the postures “11”, “**”, “21”, “32” make up 84% of the data.

Torso

For the Torso the conventional driving posture “22” and the slouching posture “32”, where the hip is moved forward from the neutral position, dominate the chosen postures. “22” makes up 77% and “32” makes up 12% of the total. The forward leaning posture “11” was chosen in 6% of the cases. When comparing “Level 3” and “Level 4” a slight increase in the postures differing from the neutral posture (“22”) can be observed. The complete frequencies for every torso posture are displayed in Table 7.

Hands

Of the 91 combinations of possible hand postures 40 appear in the dataset. Since 30 of the occurring make up less than 2% of the total, only the top ten are reported in Table 8. The most frequent postures are “1w1w” and “1r1r”. The comparison between “Level 3” and “Level 4” shows little shifts for the different conditions.

Legs

For the legs the comparison between “Level 3” and “Level 4” shows a big difference for the participant behavior. The distribution of the frequencies shifts towards postures taking up large space. In “Level 4” the frequencies are more evenly distributed. The complete frequencies for every leg posture are displayed in Table 9.

3.2 Condition and NDRT Specific Results

The level and task specific frequencies for postures and tasks are reported in the following paragraphs. Two tables are generated for each of the two conditions. One contains the frequencies of postures taken while performing a certain task. The other displays the frequencies of tasks performed while holding a certain posture. A total of four tables exist per body region. An empty cell is equal to a frequency of 0. The more empty cells are present in the table the less different postures were chosen by the participants. The Tables 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24 and 25 can be found in Sect. 6 and contain the described data. By assembling the body region postures with the highest probability the most relevant full body postures can be identified.

4.1 Data

Head

The posture of the head is mainly dominated by the objects held in the hands. The user needs visual feedback to operate. If nothing is held in the hands, the head usually moves a lot. Thus the head can be used as a predictor to decide between two states, task with object in use or task without object.

Torso

The data shows that the torso is likely to be in a normal driving state. This can be traced back to the fixed backrest, although the participants were allowed go into a slouch posture and often did not decided to do so.

Hands

The hands are the best predictor for a task when a posture is given. Similar to the head a clustering of tasks with similar properties is practicable. A difference between the two conditions is quite small. This hints that the given space at “Level 3” is already sufficient for most of the tasks. Working with the Laptop needs more space, because of size of the used object. The study did not include any supporting elements for the arms. This has a major influence on the posture as humans tend to adopt to their environment.

Legs

The posture of the legs does not allow any conclusion about what task is executed. From the opposite perspective human factors engineers or designers need to mind, that the users generally may take every leg position imaginable if they have enough space. This is the case for both of the conditions. As it is shown in the data the pedals in “Level 3” hinder the more extended postures.

4.2 Methodology

This study was conducted with 25 participants. Thus a descriptive statistic approach was used to identify big effects of the automation conditions. If done with a larger population the inferential statistics might be applicable. To keep the complexity of the experiment at a manageable level the backrest was fixed. This needs to be researched in future studies to understand the full scope of body positions in relation to NDRTs. The same applies to anthropology, time, vehicle dynamics and additional interior elements (e.g. footwell dimensions, doors or center tunnel). The coding system needs to be used by hand thus making data assessment time intensive. Depending on the research aim it might also be beneficial to make it more or less precise.

Being relevant for safety, comfort and ergonomic assessments this data has to be handled with care. The weighting of the tasks was kept equal here, but can have differing relevance depending on the use case of the car. Other studies show the magnitude of this factor [10,11,12].

This study is among the first steps of the research to find the postures of drivers in automated cars, for designing the drivers place of future automated vehicles This is necessary since the state of the art methods of designing car interiors will find their limits in future applications.

Two studies need to be conducted in the near future. One to find preferred backrest angles of passengers during NDRTs and the other to make the connection between needed space and NDRTs.

Head

Torso

Hands

Legs