30.11.2012 Views

Automotive User Interfaces and Interactive Vehicular Applications

Automotive User Interfaces and Interactive Vehicular Applications

Automotive User Interfaces and Interactive Vehicular Applications

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

• Telephone application<br />

– Call the contact “Staffan”<br />

– Add the contact“Elin”with home number“118118”<br />

• Logistic application<br />

– Ask the system to read the first order<br />

– Show your colleagues on the map<br />

The test leader planned the test so that the participants<br />

started each task at approximately the same spot along the<br />

route. We altered the test conditions so that one third of<br />

the participants began with the SUI condition, one third<br />

with the GUI condition <strong>and</strong> one third with the multimodal<br />

condition. This was done to minimize the bias from learning<br />

the system during the test so that the last condition is<br />

the easiest to use. Before starting the car, the test leader<br />

presented the tasks <strong>and</strong> demonstrated the first task (call<br />

the contact “Staffan”) by using the condition that the participant<br />

should start with. Before starting a new condition<br />

the participant was told to stop the car so that the test<br />

leader could demonstrate that condition. The participants<br />

also practised the task while the car was st<strong>and</strong>ing still.<br />

5. RESULTS<br />

We wanted to compare ease of use <strong>and</strong> how the driving ability<br />

changed depending on modality. Task performance was<br />

measured by the test leader on a 4-grade scale:<br />

• Ok without support<br />

– the task was performed without any help from the<br />

test leader<br />

• Ok with light support<br />

– the test leader gave one or two instructions<br />

• Ok with considerable support<br />

– the test leader gave three or more instructions<br />

• Not ok, task is aborted<br />

– the test leader or the participant decided to abort<br />

the task<br />

• Ok with slight mistake<br />

– the task was performed, although there were a<br />

slight misunderst<strong>and</strong>ing of the task. For example,<br />

the participant added the wrong phone number or<br />

called wrong contact. In these cases, the mistakes<br />

were due to wrong input from the participant <strong>and</strong><br />

not a misinterpretation from the system.<br />

Driving ability was estimated by the participants. After<br />

each task, the participant was asked to estimate the driving<br />

ability during the task on a 10-grade Likert scale, where 10<br />

meant “My driving ability was equally good as when I drove<br />

without performing any task” <strong>and</strong> 1 meant “My driving ability<br />

was really bad”.<br />

We also measured task completion time for each task, starting<br />

with the participant pushing the first button to either<br />

start the voice recognition or choose menu on the manual<br />

interface, <strong>and</strong> ending with the participant returning to the<br />

main menu.<br />

5.1 Task performance<br />

Figure 5.1 shows task performance.<br />

("<br />

!#'"<br />

!#&"<br />

!#%"<br />

!#$"<br />

!"<br />

)*+" ,*+" --"<br />

./"0123"451632"71428/9"<br />

:;2";/8=59"<br />

4BCC;>2"<br />

./"0123"51632"4BCC;>2"<br />

./"0123;B2"4BCC;>2"<br />

When using the manual interface, the participants did not<br />

need much help from the test leader, <strong>and</strong> the tasks were<br />

performed with no or only slight support.<br />

When using the speech interface, the majority of the tasks<br />

were performed with no or only slight support from the test<br />

leader, or was performed ok but with a slight mistake. Approximately<br />

20% of the tasks needed more support or was<br />

aborted. In the cases where the task was aborted, all in<br />

all three times, it was due to the system not being able to<br />

perceive the input or misinterpreting the input.<br />

Multimodal interaction was similar to GUI interaction, although<br />

a few tasks needed considerable support, or were ok<br />

but with a slight mistake. In one case the task was aborted;<br />

this was due to the user using an expression that was not<br />

covered by the lexicon. Instead of using the keypad, the<br />

user chose to abort <strong>and</strong> then restart the task <strong>and</strong> was able<br />

to perform it using another expression.<br />

5.2 Self-estimated driving ability<br />

Table 1 shows self-estimated driving ability.<br />

Modality SUI Multimodal GUI<br />

Mean 7,5 7,25 5,48<br />

STD 1,90 1,76 1,91<br />

Median 8 7,5 6<br />

Table 1: Estimated driving ability according to the<br />

drivers.<br />

The participants estimated their driving ability higher when<br />

using the SUI or the multimodal interface, compared to the<br />

GUI interface. A one-way Anova test shows that the difference<br />

between the GUI compared to the SUI <strong>and</strong> the multimodal<br />

interfaces is statistically significant, F(2,113)=12,923,<br />

p

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!