2.3.1. CARACTERIZACIÓN TEJIDO Y FIBRA DE CABUYÁ
2.3.1.5. ENSAYO DE TRACCIÓN EN TEJIDOS (ASTM D5034-ADAPTACIÓN)
In several workshops with HCI experts we gathered and discussed a large number of measures that could be used to support the recognition of automation error by means of HCI. Below we present a selection of the discussed approaches and a common judgement by assumed effectiveness, acceptance and feasibility.
One of the simplest, but probably most effective measures, is for the driver to keep the
hands on the steering wheel. Even with automated longitudinal and lateral control, with no
immediate necessity to have the hands on the steering wheel, the haptic feedback from the road through the steering wheel can provide helpful information about the vehicle dynamics and therefore a better connection to the driving task (cf. [27]). Some internal studies have been conducted which indicate high effectiveness of this measure. Even with the eyes off the road, the driver is still aware of his or her situation in a moving vehicle on the road. So the supporting role of this measure is achieved by the maintaining of driving-related feedback. Driver acceptance is questionable, though, since a potential gain of comfort might be lost if the driver cannot take the hands from the steering wheel. However, we require "hands-on driving" in all user studies presented in this work.
Another potential measure related to the one named above is the provision of vestibular
feedback, i.e. addressing the vestibular organ by deliberately changing speed or direction to a
noticeable extent. Of course, this is only possible on a very small scale in real traffic and driver acceptance is doubtful. Moreover, sudden acceleration, braking or steering without relation to outside events can be counter-productive and might be interpreted as a system malfunction.
Targeting the engagement in secondary tasks during automated driving, in particular visu- ally distracting tasks presented on an in-car display, discussions about deliberately directing the driver’s attention arose. In order to avoid unintentional distraction, in-car displays showing secondary task content could be turned off by default when driving automatically. Only when a
button is pushed, the display is turned on as long as the driver presses the button down. When
the button is released, the display is turned off again. Thus it is always known to the control systems in the car when the driver is likely to watch the display. It is questionable, though, if such a system design is acceptable to the driver.
Based on Lee’s proposal to takeoperation actions and gaze behaviourinto account [96], there are a number of options to react to driver gaze and input behaviour. In order to motivate the driver to check the road frequently, displays could be accessible only for a limited amount of time. After a while of watching the display, or after a certain time of input operation, it could be turned off or, in a more ambient way, subtly reduced in quality. It is likely that such measures will not find broad acceptance, since they all patronise the driver. Moreover, eye tracking systems inside the car create significant technical expenditure. It is questionable if forcing the driver to look back to the road at certain intervals is really efficient and would contribute to detect automation errors.
With the eyes already from the road, would it be possible to bring the road to where the eyes are? This could be achieved by avideo image of the driving scenepermanently visible in a splitscreen-view or displayed repeatedly in certain intervals could contribute to an increased situation awareness by having the driving scene at least in the peripheral field of view even when the eyes are not on the road (cf. Fig. 3.9). However, this bears the risk of drivers relying on the camera image alone and watching the road and the automated control systems even less. Because of the limited aperture angle of the camera, driver’s field of view is smaller than the real road view and objects beyond the field of vision cannot be seen. A variation of this could be
anabstract view of the driving sceneinstead of the original camera image in order to reduce
the details to be displayed, e.g. in the form of a differential contrast image. The danger is that critical content is lost due to the abstraction.
Figure 3.9: When engaging in a visual task on a display, a video image of the frontal driving scene could be displayed in a splitscreen-view (montage).
A promising approach appears to be the augmentation of the driving scene with helpful information, for instance the projected driving path that will be taken by the vehicle in the next seconds, based on sensor information and the computed trajectory by the automated control system. Thus the driver even gets more information about the state of the automated control system as he would by watching the road alone. The visualization of an otherwise hidden system state and the future projection of this information might bear the potential of facilitating in a unobtrusive and simple way the recognition of incorrect automated control. This idea is further pursued and elaborated in chapter 5. Different ways of realising such a system are discussed in 5.1.
3.4
Summary
This chapter presented the methodological and procedural fundament of this work. We intro- duced a comprehensive formal categorisation of secondary in-car tasks with the dimensions modality, degree of interaction, interruptibility and information encoding. This will serve as a decision aid for the selection of tasks in the following chapters. We argued for the need of a dedicated methodology designed for the assessment of error recognition in an automated driving context and described the process of finding a suitable platform. Since we do not only want to assess the driver’s ability to recognise automation errors, but also provide means to improve error recognition, we provided a discussion on potential supporting measures. Displaying the future driving path was identified as the most promising measure to support the driver in recognising automation errors.
Development of the Automated Lane
Change Test
This chapter presents the development of a novel methodology to assess human detection of errors in the context of automated driving. Section 3.2.2 has described the motivation, the back- ground and related method that have led in the end to the proposed methodology, theAutomated
Lane Change Test (ALCT). In section 4.1 we explain in detail the characteristics and qualities
of the ALCT and why they were designed that way, as well as the technical implementation. We conducted two studies using this method assessing the recognition of errors in an automated driving scenarios, and the methodology itself. Parts of this chapter have been published in [153] and [154].
4.1
System Design
As already mentioned, the ALCT method is designed as an instrument to judge the recognition of errors in an automated driving scenario. We assume a semi-automated system as a technical basis that takes over the stabilisation and partially the manoeuvring task, i.e. is able to maintain the speed and keep the vehicle inside its definite lane. Moreover, it is able to perform lane change manoeuvres autonomously.