UC Berkeley’s Institute of Transportation Studies (ITS) is proud to announce a new framework co-developed with Lawrence Berkeley National Laboratory (Berkeley Lab) and Uber that opens up new opportunities for communities to envision future transportation solutions with a new software tool called: Berkeley Integrated System for Transportation Optimization (BISTRO).
“BISTRO is an exciting new mobility pattern optimization and modeling tool in the age of data science that can help communities plan for the future as transportation needs change city landscapes,” says ITS Director Alexandre Bayen.
Using advances in machine learning and data science, BISTRO uses data-driven transportation demand modeling and policy analysis at a resolution and scale that city officials, transportation system managers, the private sector, academics, and citizens can understand and analyze while planning for the rapidly evolving transportation realities shaping urban areas worldwide.
“Uber funded research into BISTRO's development as part of its vision to foster participation and innovation in urban mobility research,” says Jonathan Lee, Senior Engineering Manager & Project Coordinator at ITS, UC Berkeley, who collaborated on the project as a Senior Data Scientist at Uber.
Berkeley researchers and Uber teamed up to address large-scale transportation planning that shows how different modes of transportation shape and change congestion.
The new analysis and evaluation platform works in concert with Behavior, Energy, Autonomy, and Mobility (BEAM) simulation software developed at Berkeley Lab to enable open-sourced development and evaluation of transportation optimization methods in response to given policy priorities.
When the idea of developing this kind of platform arose, Sidney Feygin, head of Simulation at Marain and former consultant at Uber, looked to Berkeley Lab’s BEAM, a technology he previously worked on, as the software to build this collaboration on. Feygin and a team of ITS doctoral students, Jessica Lazarus and Edward Forscher, as well as Uber’s Valentine Golfier-Vetterli, Lee and Abhishek Gupta, Berkeley Lab’s Rashid Waraich and Colin Sheppard, and Bayen teamed up to tackle the problem and created BISTRO.
Built for city planners, policy makers and transportation experts, BISTRO shows how a set of changes to existing transportation systems can bring about the greatest improvements across important indicators of transportation system performance in terms of system-wide level of service (LoS), congestion, accessibility, and sustainability.
“BISTRO can help cities reduce uncertainty about how possible future policy interventions and technology disruptions could impact long-term urban planning objectives,” says Feygin.
With an agent-based model, every person and transportation unit — personal or shared car, bus, train, etc. — is represented, allowing microscopic interactions between fleet operations, fares, and vehicle availability to accrue over time and space. The macroscopic traffic patterns emerging from these interactions affect key indicators and help leaders make more informed decisions about transportation needs.
“Agent-based models are the only elegant way to get a high resolution, high fidelity whole picture with all the players,” says Sheppard, now Chief Architect at Marain and former Computational Research Scientist at Berkeley Lab. “You can really analyze the complex relationships between the supply and demand of transportation when you are able to model every person and vehicle in a city and simulate their decision-making in how they will travel to their destination.”
During a March 2019 hackathon at Uber, the team was able to test BISTRO. As part of this pilot, Berkeley researchers developed a full-scale transportation system model compatible with BEAM that was based on the city of Sioux Falls, South Dakota. About 400 engineers participated in developing and implementing scalable global optimization algorithms aimed at improving performance measures such as environmental sustainability, accessibility, and system-wide congestion.
“It was a great learning experience and a fun opportunity for many participants to apply their skills to a problem different from their day-to-day responsibilities,” says Lee.
During the hackathon, both Berkeley researchers and Uber identified bugs in the system objective, which were corrected prior to releasing the platform for public consumption.
“Developing a system objective that truly reflects urban planner’s intent requires an iterative ‘debugging’ approach. One important outcome of the hackathon is that BISTRO is now equipped with a set of default global optimization routines, which, as a best practice, should be used to identify possible problems in the definition and/or implementation of candidate objective functions,” says Feygin.
In the past academic year, PhD student Jessica Lazarus has led a team of researchers at ITS Berkeley to continue development of BISTRO and carry out a study of congestion pricing optimization using the platform. She presented this research at the 26TH ACM SIGKDD conference on knowledge discovery and data mining.
“BISTRO enables us to investigate road pricing optimization from multiple angles – from the price structure and geographic coverage of the policy to the impacts of different optimization approaches on individual and system-level policy objectives,” says Lazarus.
The open source platform is available on the website bistro.its.berkeley.edu for the public to use.