A Brief History of Tiramisu Transit

Aaron Steinfeld, Carnegie Mellon University

When we recently shut down the public, deployed version of the Tiramisu Transit app, I sent a personal note to Kari Watkins and Sean Barbeau. We have collaborated on larger transit issues since the early days of our respective efforts. Like OneBusAway, our app has deep roots in answering important research questions on how to improve the transit experience. 

Our group at Carnegie Mellon is a research team, not a company. We decided to retire Tiramisu because multiple companies have adopted our innovations in crowdsourcing, real-time information, and universal design. For example, Google Map’s transit information now shows vehicle fullness and supports crowdsourced information – features we introduced to the public when Tiramisu launched in 2011. 

While the app launched to the public in the summer of 2011, the origins of the Tiramisu Transit line of research started in 2008. During an early lab study in the Rehabilitation Engineering Research Center on Accessible Public Transportation (RERC-APT), a participant remarked that reporting and tracking accessibility barriers was a great idea, but if the bus was too full for their wheelchair, the problem of barriers was moot. In parallel, participants who are blind noted the lack of real-time arrival information for transit users in Pittsburgh. This led to long waits outdoors in unsafe locations and during inclement weather. In the late 2000’s, such systems were cost prohibitive for many transit systems to install and there were no plans to acquire one here in Pittsburgh.

Our team observed that the combination of crowdsourcing and smartphones could gather both fullness and arrival time data. We launched an effort to design and build this kind of system where users could report fullness and trace transit vehicles while on board. The first key innovation was to bootstrap from static schedule data and layer in fullness and arrival times for both real-time (from peer phones) and historic (from observed distributions) estimates. The second innovation was to use universal design to increase value to all riders. This enabled transit riders without disabilities to (usually unknowingly) generate critical data for riders with disabilities. 

After a closed pilot test, we launched Tiramisu Transit version 1.0 to the public using a combination of the RERC-APT core funds and seed money from the newly started Traffic21 initiative. In true Pittsburgh form, riders contributed large amounts of data from all over the city and suburbs. Over the next few years we conducted various research studies and updated the system to incorporate newly available real-time data feeds (version 2.0). Research during this time showed (1) that many users benefitted from their peers’ contributions, (2) crowdsourced prediction performance was quite good, and (3) characterized how riders wanted to communicate with service providers. We also refined and characterized vehicle fullness labels, including comparisons to automatic passenger counter data. These labels are now in GTFS-Realtime. Most importantly, we documented how localized transit information removed barriers to travel; we observed participants with disabilities engaging in less preplanning and more opportunistic travel.

Science led us to change Tiramisu for version 3. We realized that arrival times and fullness ratings were becoming an implementation problem, rather than a research problem. Our experiences led us to recognize the value of personalized transit information, especially for riders who use screen readers, have cognitive disabilities, or have difficulty manipulating touchscreens. We redesigned and reimplemented the system so users could select filters on trip direction and routes. Tiramisu learned personalized models from these selections based on location, time of day, and day of the week. Our research showed a clear reduction in user effort for users (both with and without screen readers).

Finally, our teams have learned from each other for years and we encourage others to build on and expand this relationship. This goes beyond academic camaraderie and transit advocacy. In the open source code of Tiramisu version 3.0, you will see OneBusAway in the real-time transit information pipeline.

The initial ambition of Tiramisu Transit was small. We wanted to advance research knowledge in a way that would have a positive impact on transit riders with disabilities. Along the way, our work had a long-term, meaningful impact on a wide range of stakeholders. While many research projects receive validation only from only the scientific community, our proudest validation came from seeing our innovations adopted by companies for global use and the thousands of people who used our system to improve their daily lives.

Screenshots of Tiramisu Transit versions 1 (left) and 2 (right)

Screenshot of Tiramisu Transit version 3 showing manually selected filters (blue) and machine learning selected filters (orange)

Images are © Carnegie Mellon University 2022

Results from case studies of OneBusAway deployments

OTSF recently contracted with Garnet Consulting to produce a set of case studies of OneBusAway deployments, to freshen up our website, and to write a one-page flyer that can be distributed to prospective users and at conferences. The flyer is now linked from the OneBusAway website, as are capsule summaries of the case studies. We also have interview transcripts and other details from the case studies available as needed.

The interviews confirmed that our users appreciate that OneBusAway is an open-source platform and that their data belongs to them. One interviewee said “When I hear OBA I think open source.” Another said they picked OneBusAway because “We wanted control of our data. We wanted to be able to leverage whatever we did without running into ‘who owns it’. We’ve been able to do what we want.” Using OneBusAway allows agencies to make decisions centered around their specific needs, without focusing on vendor restrictions. Interviewees also voiced appreciation for OneBusAway’s commitment to accessibility. One interviewee described that OneBusAway has “always put accessible information front and center.” Another user reported that blind riders on their internal committee “recommend the app to people because it works for them.”

Another takeaway from the interviews was that some OneBusAway users initially had confusion about what options were available for setting up OneBusAway for their agencies. To begin addressing this, we’ve made some updates to our website to make it clearer how to get started with OBA. We’ve also added case studies and a list of OneBusAway deployments to the site so that visitors can see examples of how different agencies are using and benefitting from OneBusAway.