Multimodal Transit Network Design with SAVs

City-scale multi-commodity network flow and hybrid metaheuristic/NLP optimization for integrating Shared Autonomous Mobility Services (SAMS) with public transit in collaboration with Argonne National Laboratory.

Emerging Shared Autonomous Mobility Services (SAMS) and Shared Autonomous Vehicles (SAVs) present transformative opportunities to reinvent public transit systems. While uncoordinated autonomous on-demand fleets risk cannibalizing high-capacity public transit ridership and worsening urban congestion, coordinating SAVs as first-mile/last-mile feeders to high-frequency transit backbones can unlock unprecedented accessibility and operational efficiency.

In collaboration with researchers at Argonne National Laboratory, this project develops city-scale mathematical programming models, continuous approximations, and hybrid metaheuristic optimization algorithms to jointly design multimodal transit service networks, vehicle stopping patterns, line frequencies, and autonomous vehicle fleet sizes.

System architecture for large-scale multimodal transit network redesign, integrating continuous approximation of feeder zones with trunk network design and point-to-point SAMS operations.

Methodology

The research develops a tiered optimization and analytical modeling framework designed for computational scalability on metropolitan-scale networks:

  1. Continuous Approximation (CA) for Spatial Demand & Feeders: Characterizes regional geography, trip densities, and localized feeder operations analytically. This captures door-to-door access, waiting, transfer, and in-vehicle journey times across heterogeneous urban and suburban zones without requiring computationally prohibitive agent-based simulation loops during strategic planning.

  2. Destination-Labeled Multi-Commodity Network Flow (MILP MCNF): Formulates the urban transit pattern design problem as a mixed-integer linear program (MILP) tracking destination-specific passenger flows. The model simultaneously determines stop sequences, express/local stopping strategies, short-turning turnarounds, and deadheading trips across interconnected transit corridors, accounting for perceived passenger headways and cross-pattern transfers.

  3. Hybrid Metaheuristic & Nonlinear Programming (PSO + NLP): Solves the joint multimodal transit frequency setting and SAV fleet allocation problem. A Particle Swarm Optimization (PSO) metaheuristic explores combinatorial line frequency spaces and strategic route pruning, while Nonlinear Programming (NLP) locally optimizes continuous fleet deployments. The framework incorporates multinomial logit (MNL) mode choice (transit, point-to-point SAV, private driving) and analytical vehicle utilization models.

Left: Optimization convergence and mode share reallocation under hybrid PSO-NLP frequency setting. Right: Destination-labeled multi-commodity network flow patterns optimizing stop skipping and line transfers.

Key Findings & Case Studies

The optimization framework was tested and validated on the Chicago Metropolitan Area multimodal transit system, encompassing CTA rapid transit (“L”), CTA bus lines, Metra commuter rail, and suburban Pace bus systems:

  • +33.3% Transit Ridership Growth: By redeploying low-frequency, winding suburban bus services into agile SAV feeder zones that connect to rapid rail trunk corridors, regional transit ridership increased by 33.3%.
  • Off-Peak Accessibility Boost: Dynamic SAV feeder integration dramatically mitigated off-peak service degradation, boosting suburban transit connectivity during evening and weekend travel windows.
  • Journey Time & Fleet Savings: Coordinated express/local pattern optimization and short-turning reduced total passenger weighted journey times by up to 4.13% across major metro lines, while lowering operating vehicle requirements and capital costs for the public transit authority.

References

2026

  1. Joint optimization of multimodal transit frequency and shared autonomous vehicle fleet size with hybrid metaheuristic and nonlinear programming
    Joint optimization of multimodal transit frequency and shared autonomous vehicle fleet size with hybrid metaheuristic and nonlinear programming
    TR Part C Transportation Research Part C: Emerging Technologies, Apr 2026

2025

  1. Joint Optimization of Pattern, Headway, and Fleet Size of Multiple Urban Transit Lines with Perceived Headway Consideration and Passenger Flow Allocation
    Joint Optimization of Pattern, Headway, and Fleet Size of Multiple Urban Transit Lines with Perceived Headway Consideration and Passenger Flow Allocation
    TRB In 104th Transportation Research Board Annual Meeting, Apr 2025

2024

  1. Redesigning large-scale multimodal transit networks with shared autonomous mobility services
    Redesigning large-scale multimodal transit networks with shared autonomous mobility services
    TR Part C Transportation Research Part C: Emerging Technologies, Mar 2024