For our final project in User Interface Design at UC Berkeley's School of Information, we redesigned the Bay Area Rapid Transit (BART) ticket kiosk. Our goal was to improve the user interface software within the limitations of the existing 8-button, ATM-style physical interface - instead of going with a touch-screen. We believe our design offers major usability advantages, especially for infrequent riders. Pick the location of the kiosk and give it a try. Let us know what you think at
We chose to retain the existing BART ticket kiosk hardware, specifically the 12-inch display and eight-button physical interface. We had considered doing away with the buttons altogether and designing for a touch display, which would have opened up any number of design possibilities. However, we concluded that developing new software for the existing hardware would make the prototype much less expensive for BART to implement (if they ever chose to).
This decision has yielded exciting improvements. For example, users can purchase tickets to destinations they select on-screen instead of having to look up fares on a separate list. Riders can also buy tickets to either Bay Area airport with only two button presses. Preliminary user studies show that riders can buy tickets much more quickly with our prototype than with the existing kiosks. More more details, read the final report on the redesign.
We constructed a physical replica of the BART kiosk user physical interface using an Arduino micro-controller to relay button presses to the software built in Flex 3 via USB. The replica slips over a laptop display mimicking the appearance of the BART kiosks. We used an open source project called AS3Glue to register serial communications from Arduino and dispatch ActionScript events to the software.
Ljuba looks on as a BART rider uses the prototype.
At this time, the prototype simulates transactions made by credit or debit card only. This limitation was imposed for practical reasons: we could not build a physical prototype capable of accepting cash or existing tickets as payment. Simulating credit card transactions is much simpler and sufficient to demonstrate the effectiveness of our prototype as a viable alternative to the existing kiosks.
Second, the experience of using the software interface above is significantly diminished without the affordances of the physical prototype. Being at a BART station - with the physical kiosk in front of you - reduces the perception that the screen is touch-sensitive. Also, the metal bezel and buttons support the sense that the buttons are connected to words next to them on-screen.
Ben Cohen and Ljuba Miljkovic are Master's students at the UC Berkeley School of Information, finishing up their first year; we don't work for BART. Ben's focus is on user interface design and usability. Ljuba's interests are in user interface, product and interaction design.
To buy a new ticket, a user inserts either cash or a credit/debit card into the machine, then chooses to either print a ticket of the value inserted ($20 for credit/debit) or subtract value in $1 or $0.05 increments to adjust the value of the ticket. Fares from the current station to all other stations are published separately on the kiosk.
We used a mix of qualitative and quantitative methods to test our prototype with actual BART riders. We wanted to see whether it would yield faster transaction times and whether users would feel more satisfied with the ticket-buying experience overall. The testing took place over two days in the Downtown Berkeley station.
Here are representative use-cases for someone familiar with BART and new to our prototype:
Without any instructions, this rider quickly discovered what she needs to do to buy her ticket with the prototype.
Even though it was her first time using our prototype, it took her about as long to use as the existing kiosk when you include the time to look up the fare (not shown in the video). Her second time using the prototype was much faster. In the end, she strongly preferred it to the existing kiosk interface.
First-time and expert users of both systems performed significantly better on the prototype.