CHI2013 Rebuttal

We thank reviewers for their quality reviews and their positivism toward this paper. We can confidently address their concerns in a final revision to make our contribution even stronger.

Related Work (AC, R1)

We agree with R1 on the relevance of the work on navigational support for the visual impaired in relation to our own, and we have tried to refer the reader to such venues in our introduction. However, the limited space available for a tech-note limited our ability to extend our description of salient prior art into a full-fledged related work section. Vibro-tactile and audio feedback are ideal alternatives for visually impaired users. We considered both these alternatives for our design, but decided not to use them. While vibro-tactile feedback can be useful and can even encode multiple dimensions, further research is needed to determine proper mappings of depth information to tactile stimuli. Moreover, Brewster et al. showed that when in-the-wild "the variations introduced by the environment mask the small benefits found in the lab" [a], thus minimizing the impact of tactile feedback while on-the-go, which is a primary aim of our approach. Moreover, audio feedback requires users to either use a headphone or walk in areas with relatively minimal ambient noise, which is not our case as we aimed to support walking in busy places. This analysis reinforced our design choice of using visual feedback, for this first investigation. We will add a short description of this rationale in the design section in the final revision.

[a] - Tactile feedback for mobile interactions - http://dl.acm.org/citation.cfm?id=1240649

Presentation (AC, R1, R2)

For the final version we will optimize the text to remove some of the repetition (particularly the brief reporting of the results in the intro as is common in a full paper), opening up this way some room for the new content on related work and design rationale as discussed above.

Evaluation (R1)

For the evaluation we considered other types of applications including texting (R1). However we decided for the game because text-entry does not really demand from users to stay focused on the display while texting; that is, the SMS task can be interrupted any time without any cost for the participant (error rate). While it is true that the game demands less cognitive resources than actual composing and sending of a SMS, the game introduces an element of pressure on the participants to stay focused and maximize their viewing of the display instead of their surroundings, incurring a cost when they lose focus of the screen. This canonical task was selected primarily to ensure that users are mostly focusing on the display, as not doing so could confound our results.

R1 correctly points out that users could divert their attention to the ambient band instead of raising their head to find their way, especially for the Color condition. This situation is actually desirable (we designed for it) and was actually expressed by users in our interviews, as we reported in the results section. Users do look at the ambient band; however glancing at the band (switch of focus on the display) is simpler than raising the head. Moreover, results showed that users preferred the low fidelity images (depth) because they could gather information faster (less cognitively demanding). Finally, the apparent increase in the number of near collisions (R1) did not yield a significant difference, and that's why we did not elaborate on that outcome in the paper. We will make it explicit in the final revision that such a difference was not significant.

Other

• R1, we envision a cellphone design where the depth camera can re-align to the horizon (and thus capture whatever obstacles are on the user’s path) pretty much the same way a gyro horizon works on an airplane (battery-less). We also envision CrashAlert as a OS service showing the ambient band on top of the actual application, so that application developers do not have to worry about resizing their apps.
• R2, we appreciate your remark on off-screen visualizations (e.g. halo) and we will add a citation to it in our future work.
• R2, we normalized the handling manoeuvres to 100% for the sake of the visualization. However we analyzed the data on the raw numbers we collected.
• R3, the actors caused only 4 potential collisions per trial. As we report in the paper, a trial had 11.26 (sd = 2.96) potential collisions on average.

We will carefully revise the paper to include the above discussions. We also thank reviewers for other (minor) suggestions which we will apply in a final revision.