To date, safety services are typically constructed as dedicated stovepipe systems focusing on high reliability and a specific area of risk (e.g., automotive safety systems). Usage of such services remains limited since they require a dedicated investment for each system. This project trades off the ultra-high reliability of dedicated systems for the much more rapid adoption of safety services that comes with integrating them directly into mobiles and wearables. By demonstrating the feasibility of this approach, this project can contribute to saving lives, such as some of the more than 30,000 traffic fatalities in the United States each year. It can also inform regulatory policy for safety services at the CPSC, NHTSA, or FCC. Moreover, the PIs will not only train graduate students to conduct the research but also actively include undergraduates and high school students through research internship programs. Results will be disseminated through scholarly publications, active outreach to the wireless and mobile industry through WINLAB's industry events and connections.
This project seeks to demonstrate that the mobile devices we carry and wear can provide effective safety services. This is particularly relevant where our devices contribute to dangers by causing distractions for drivers and pedestrians. This project therefore pursues the vision of a system that offsets such unsafe use by continually sensing our activities and surroundings, identifying potentially dangerous situations, and mitigating them through appropriate interventions. At a technical level, the primary challenge lies not only in designing precise sensing techniques but in understanding and managing the level of confidence provided by these techniques. A key observation is that there are usually multiple possible interventions of varying levels of intrusiveness and tolerance to false positives. It is therefore important to match interventions to the confidence level provided by the sensors. To address this challenge, the project develops system support and a toolkit to help developers track and manage mobile sensing uncertainty. It explores crowdsourcing failure and relevance data from a large user population and automatically estimating the confidence provided by internal sensing and activity recognition components. The toolkit can further use the obtained metrics to help adapt sensing or application behavior. The system might conserve energy by switching one context sensor to a fallback mode from a diversity mode; or, the system could switch to a different intervention if the level of confidence has changed. System validation includes prototyping two application use cases, which sense and mitigate mobile device distractions for drivers and pedestrians. Together, these techniques form the system, which supports development of many other effective safety services on mobile devices.
