Final Task Categorisation

For a final categorisation we took the previously reported work into account. Wickens’ model of multiple resources ( [172], as described in 2.1.2) turned out to be most helpful. Wickens categorises the available resources into modality (visual and auditory information) and encoding (verbal and spatial). Also in ADAM tasks were judged by visual and auditory information. For supervising an automated control system, mainly visual resources are needed. In case of a secondary task also requiring visual resources, the resource conflict will cause interference

resulting in a performance decrease in the primary task [151]. Verbal information can be conveyed in writing or orally, whereas spatial information in our definition means information that requires visual thinking and the capacity of thinking in two or three dimensions, independent from a vocal or manual response.

We adopted Wickens’ resource dimensions modality and information encoding for the categorisation of arbitrary secondary tasks that are likely to be performed during driving in a car. We decided not to include ’haptic’ or ’kinaesthetic’ as modality dimension, because we could not find a common secondary in-car task that solely conveys information through a haptic or kinaesthetic channel. Since modality and information encoding did not suffice for a distinct categorisation, we included another dimension regarding thedegree of interactionnecessary to perform a task. Studies indicate that the degree of interaction plays an important role in driver distraction (e.g. [157]). ADAM additionally distinguished motoric and cognitive demand. Since our focus is on tasks that require technical means, and therefore could potentially be controlled by the car manufacturer, those demands often go hand in hand. So we decided to subsume these toactiveandpassive tasks. If an active engagement is necessary, such as operating a character input device or talking on the phone, we term this as an active task, whereas a task with a more consuming character, such as reading or listening, is categorised as passive. An active task can probably also be described as a task that induces high mental workload.

We also tried to account for guidelines issued by official authorities. As mentioned, the AAM for instance demands easy interruptibility of in-vehicle systems. This means that the perceived cost to interrupt and resume an ongoing task must be as low as possible. This includes visual processing, visual search, absence of external memory aids, temporal pressure, loss of coherence, and motivation [1]. The criterion of interruptibility was judged by a number of experts by the duration of a minimal subtask (operating action, written or spoken sentence, etc.). For instance, reading a news headline consisting of a few words is rated as better interruptible than reading a fully spelled sentence. The final categorisation scheme consists of the dimensions:

– Modality: visual – auditory,

– Degree of Interaction: active – passive,

– Interruptibility: good interruptibility – bad interruptibility,

– Information Encoding: verbal – spatial.

Out of consistency reasons, we chose a dichotomous (two-part) scale for each dimension, in par- ticular for the degree of interaction and interruptibility, i.e. each dimension has only two extreme factor steps. It would have been easy to introduce more than the two levels in each dimension, but since the dimensions adopted from Wickens (modality and information encoding) only have two possible levels, we chose a dichotomous scale for all dimensions. According to our definition, the driving task can be categorised as follows:

– Modality: Visual. As mentioned in section 2.1.1, driving-related information is conveyed

visually for the most part. Driving without the sense of hearing is probably less dangerous than driving without sight.

– Degree of Interaction: Active. The driver is actively in the control loop, must take care of stabilisation, manoeuvring, and navigation.

– Interruptibility: Bad interruptibility. Permanent action is necessary.

– Information Encoding: Spatial. The vehicle moves in a three-dimensional environment.

This is true for both manual and automated driving. Although automated driving may convey the impression of being a passive activity since stabilisation and manoeuvring is carried out by the automated system, with imperfect automation active and permanent supervising is necessary, and is therefore categorised as active and with bad interruptibility. So one of the two tasks simultaneously performed is always the driving task, respectively the task of supervising the automated driving system.

Category Modality Degree of

Interaction Interruptibility

Information encoding

1 visual active good verbal

2 visual active good spatial

3 visual active bad verbal

4 visual active bad spatial

5 visual passive good verbal

6 visual passive good spatial

7 visual passive bad verbal

8 visual passive bad spatial

9 auditory active good verbal 10 auditory active bad verbal

11 auditory passive good verbal 12 auditory passive bad verbal

Driving Task visual active bad spatial

Table 3.3: Categorisation of secondary tasks according to the dimensions modality, degree of interaction, interruptibility and information encoding.

Table 3.3 shows a number of exemplary side-tasks that are popular, likely or even designed to be performed during driving, sorted into the classification scheme by their corresponding dimensions. Secondary tasks demanding resources which interfere with the driving task are depicted with a dark background; complementary, i.e., non-overlapping resources are shown with a light background. We assume that the more interfering dimensions a secondary task

covers, the more the error recognition performance will decrease. Categories 1 to 8 describe tasks that need visual attention. Category 1 covers a classic example, the input of a destination in an integrated navigation system, that can be described as [visual – active – good interruptibility – verbal]. Good interruptibility is a key demand for those systems (cf. [1]) that are to be operated while driving, even if it is explicitly not recommended for safety reasons, and they are explicitly designed as unambiguously interruptible. This category also covers a lot of similar tasks provided by an IVIS that requires the input of characters (search for contact in the phonebook, selection of mp3, etc.). Category 2 represents the same class of tasks, but uses spatial information encoding. In-car tasks with mainly spatial information conveyance are actually rare, examples are navigation hints and interactive maps, offering the user a paper-map-like option to select a destination directly by pointing at it on a schematic depiction of the target area. Not every category has an example task assigned (cf. category 3, 4 and 8), since in some cases there were no common in-car tasks covering the corresponding dimensions. Category 5 comprises well interruptible reading tasks, such as news headlines or short text messages, as opposed to category 7 that covers less interruptible reading tasks, e.g. a long news article. Tasks in the categories 9 to 12 use the auditory channel involving active speaking (speech commands, talking on the phone) or passive listening (music, news) that also comprise different interruptibility. Auditory tasks use only verbal information encoding.