BO4Mob: Bayesian Optimization Benchmarks for High-Dimensional Urban Mobility Problem

TL;DR

BO4Mob presents a realistic benchmark for high-dimensional Bayesian optimization in urban mobility OD estimation, leveraging five city-network subnetworks and SUMO simulations to map OD demands to observed link counts. By integrating ground-truth PeMS data with stochastic, non-differentiable simulators, the paper demonstrates that BO methods—particularly TuRBO—achieve substantial NRMSE reductions over baselines, while highlighting scalability challenges on the Full Region. The benchmark supports reproducibility through open-source code and data, and it motivates extensions like uncertainty quantification, physics-informed priors, and multi‑output designs for dynamic OD estimation. Overall, BO4Mob provides a valuable testbed linking BO advances to impactful, data-driven urban mobility modeling and digital-twin applications.

Abstract

We introduce \textbf{BO4Mob}, a new benchmark framework for high-dimensional Bayesian Optimization (BO), driven by the challenge of origin-destination (OD) travel demand estimation in large urban road networks. Estimating OD travel demand from limited traffic sensor data is a difficult inverse optimization problem, particularly in real-world, large-scale transportation networks. This problem involves optimizing over high-dimensional continuous spaces where each objective evaluation is computationally expensive, stochastic, and non-differentiable. BO4Mob comprises five scenarios based on real-world San Jose, CA road networks, with input dimensions scaling up to 10,100. These scenarios utilize high-resolution, open-source traffic simulations that incorporate realistic nonlinear and stochastic dynamics. We demonstrate the benchmark's utility by evaluating five optimization methods: three state-of-the-art BO algorithms and two non-BO baselines. This benchmark is designed to support both the development of scalable optimization algorithms and their application for the design of data-driven urban mobility models, including high-resolution digital twins of metropolitan road networks. Code and documentation are available at https://github.com/UMN-Choi-Lab/BO4Mob.

Figures (11)

• Figure 1: Illustration of the overall workflow for OD estimation using BO, where sensor data and traffic simulation are used to evaluate the quality of candidate OD demands and guide the optimization process.
• Figure 2: Freeways extracted from the San Francisco Bay Area network, including a total of eight freeways. To evaluate the performance of BO algorithms, five subnetworks are utilized: (a) Simple Ramp, (b) One-Way Corridor, (c) Junction, (d) Small Region, and (e) Full Region.
• Figure 3: Convergence behavior across optimization methods. NRMSE is plotted on a logarithmic scale. Solid lines represent the mean, with error bars denoting one standard deviation over multiple runs.
• Figure 4: Fit to GT link traffic counts for the Small Region network. Each black dot represents the mean of GT link traffic counts corresponding to the OD that yielded the minimum loss for a given random seed. Error bars denote one standard deviation across seeds.
• Figure 5: TAZ configuration example with source and sink nodes.