BISTRO: Berkeley Integrated System for Transportation Optimization

• DOI: 10.48550/arxiv.1908.03821
• Publication Date: 2019-01-01
• Authors: Sidney Feygin; Jessica Lazarus; Edward H. Forscher; Valentine Golfier...; Jonathan Lee; Abhishek Gupta; Rashid A. Waraich; Colin Sheppard; Alexandre M. Bayen
• Keywords: Benchmark (surveying)

This article introduces BISTRO, a new open source transportation planning decision support system that uses an agent-based simulation and optimization approach to anticipate develop adaptive plans for possible technological disruptions growth scenarios. The framework was evaluated in the context of machine learning competition hosted within Uber Technologies, Inc., which over 400 engineers data scientists participated. For purposes this competition, benchmark model, based on city Sioux Falls, South Dakota, adapted BISTRO framework. An important finding study spite rigorous analysis testing done prior two top-scoring teams discovered unbounded region search space, rendering solutions largely uninterpretable decision-support. On other hand, follow-on aimed fix objective function, served demonstrate BISTRO's utility as human-in-the-loop cyberphysical system: one scenario-based algorithms feedback mechanism assist urban planners with iteratively refining function constraints specification intervention strategies such portfolio strategy alternatives eventually chosen achieves high-level regional goals developed through participatory stakeholder engagement practices.