Functionality

Interaction

The ALCT is set in a virtual driving scene that is identical to the scene in the LCT (straight three- lane track, no traffic, road signs indicate lane changes). However, the driving task has been fully adapted to the paradigm of automated driving, i.e. the virtual car has automated longitudinal and lateral control:

– The virtual vehicle accelerates automatically up to 60 km/h and keeps the speed perma- nently during the run.

– The vehicle is automatically kept in the middle of the lane.

– When a road sign is approached and the sign contents become visible, the virtual car starts to change to the indicated lane automatically.

As long as the automation works perfectly, no interaction is necessary. But targeting imperfect vehicle guidance, intervention is required in case of automation failure. Automation failure in the ALCT only occur in the lateral direction (cf. 1.2) and is implemented in the form of incorrect lane change manoeuvres. Following the – slightly adapted – principle of commission and omission errors (cf. 2.3), the following errors can occur (cf. Fig. 4.1):

– Lane change when the signs indicate straight driving (error 1), commission error. – No lane change when the signs indicate a change (error 2), omission error.

– Lane change indicated and carried out, but to the incorrect lane (error 3), commission error.

Figure 4.1: Error categories. Left: Lane change when not supposed to (error 1); middle: no lane change when supposed to (error 2); right: change to the wrong lane (error 3).

Naturally, error 1 can only occur when the road signs indicate to drive straight, i.e. the signs show the same lane the car already drives in. Error 2 is only possible in case of a supposed change, i.e. the signs point to a different lane the car currently drives in. Error 3 is basically a special case of error 1 that can only happen when driving on the centre lane. Errors can only occur at the point of time when the road signs become visible. Table 4.1 shows an overview of all possible manoeuvre and error constellations.

In an ALCT session, the driver is instructed to intervene as fast as possible when an au- tomation error occurs, and to correct the vehicle’s incorrect behaviour by turning the steering wheel in the direction of the correct lane, as indicated by the road signs. For instance, if the driver is on the centre lane, the road signs indicate a change to the left and the car changes to the right lane, the driver has to turn the steering wheel to the left, because that is the direction of the indicated (correct) lane. Errors can always be detected directly at sign appearance, as the car will immediately start changing lanes. Changing across two lanes – as is common in the LCT – is not allowed, since then it would not be possible to determine a clear point of time when

Target lane in case of

ID Manoeuvre Correct

behaviour Error 1 Error 2 Error 3

1 Left to centre Centre n/a Left n/a 2 Left straight Left Centre n/a n/a 3 Centre to left Left n/a Centre Right 4 Centre straight Centre Left or Right n/a n/a 5 Centre to right Right n/a Centre Left 6 Right to centre Centre n/a Right n/a 7 Right straight Right Centre n/a n/a

Table 4.1: Possible manoeuvres in the ALCT and the corresponding target lanes in case of error category 1, 2 or 3. n/a means that this error is not applicable for this manoeuvre.

an error can be recognised as such, e.g. when a change across two lanes is required and the car only changes to the next lane. In terms of the categorisation of the driving task, the recognition of automation errors happens on the knowledge-based level, the intervention with the steering wheel on the rule-based level (cf. section 2.1.1).

The characteristics of the steering wheel have been realised in two versions:

– In theLO version(low), the steering wheel serves only as a binary input device. It remains in neutral position during the drive and all lane changes. In case of an error, the driver turns the steering wheel at least 90 degrees in the correct direction and back to the original position as described above. Turning the steering wheel has no influence on the car’s driving path. An acoustical signal indicates that the driver’s action has been registered. The LO version is examined in the experiments described in 4.2 and 4.3, the implementation is described in 4.1.2.

– In theHI version(high), the steering wheel actually controls the car on the track, but the steering action is automated. Hence, the steering wheel moves according to the path of the virtual car, as it would in an actual car with automated lateral control. Due to this direct haptic feedback, the driver receives useful information about the vehicle’s trajectory. In case of an error, the driver overrules the steering wheel by applying force greater than 3.0 Nm into the desired direction. By doing so the automation is switched off and the driver must steer the car into the correct lane by himself. After the lane change is completed the automation is faded in again. The HI version is examined in the experiment described in 4.3, the implementation is described in 4.1.2.

Automation Reliability and Error Rate

A crucial question in the design of the ALCT method is in which frequency errors occur. Buld et al. recommend an overall system reliability of at least 90% in order to establish a sufficient level of automation trust [26]. As described in chapter 2.3, lower reliability and higher error rates respectively can lead to less trust in automation and therefore lower attendance to secondary activities along with lower system acceptance. Lower error rates, however, are likely to result in a number of measured values that is insufficient for statistical analysis. Following this, nine out of ten lane changes in the ALCT setup are performed correctly and one out of ten is incorrect. In total, a run comprises 90 lane changes with nine errors. Errors are randomly distributed over the course and can occur at any sign except for the first two and the last sign, as well as for only two consecutive errors in a row. Each error type (cf. Fig. 4.1) occurs exactly three times.