Analytical modeling, reinforcement learning zonal dispatching control (PPO), and agent-based simulation (FleetPy) of hybrid fixed/flexible-route SAV transit services in Chicago and Munich.
First- and last-mile connectivity remains one of the most stubborn bottlenecks in public transportation. Fixed-route buses in low-density suburban neighborhoods often suffer from poor frequency, long passenger walking distances, and high operating deficits, while fully on-demand microtransit can struggle with fleet scaling and low passenger pooling rates.
This project introduces Semi-on-Demand (SoD) transit service—an innovative hybrid directional transit paradigm tailored for Shared Autonomous Vehicles (SAVs). By uniting the schedule dependability of fixed-route transit with the spatial agility of flexible-route demand-responsive transport, SoD offers a high-performance, cost-effective feeder solution for modern cities.
Comparison of transit route paradigms along a corridor with increasing urban density: (top) Fixed route with scheduled stops, (middle) On-demand flexible route serving door-to-door/corner-to-corner requests, and (bottom) Semi-on-Demand (SoD) hybrid route combining a flexible collection/distribution portion ($x^f$) in lower-density catchment areas with a fixed-route trunk portion connecting to the transit station.
Core Methodology
The research bridges closed-form analytical optimization, agent-based micro-simulation, and artificial intelligence for autonomous fleet operations:
Strategic & Tactical Analytical Formulations: Closed-form continuous approximation models capture passenger access, waiting, and in-vehicle times alongside distance-based operating and time-based capital costs. Strategic and tactical formulations optimize flexible-to-fixed route thresholds, service headways, fleet size, and vehicle capacity. The models demonstrate that the lower operational cost structure of autonomous fleets fundamentally alters optimal transit geometry, making smaller vehicles with flexible feeder coverage economically viable.
Analytical regime boundaries characterizing when fixed, flexible, and semi-on-demand hybrid routes are optimal as a function of passenger volume per vehicle ($H\Lambda$) and relative access-to-detour ratio.
Agent-Based Simulation with FleetPy: To capture microscopic vehicle-passenger interactions, dispatch delays, and stochastic booking dynamics, I developed and contributed the Semi-on-Demand (SoD) simulation and fleet control module (SoDLineFleetcontrol) to FleetPy, the open-source agent-based fleet simulation platform maintained by the Chair of Traffic Engineering and Control at the Technical University of Munich (TUM-VT).
Deep Reinforcement Learning (PPO) Zonal Dispatching: To handle real-time demand imbalances across complex service areas, a deep reinforcement learning policy based on Proximal Policy Optimization (PPO) was designed. The RL agent observes real-time vehicle positions, passenger booking queues, and expected waiting times to dynamically route and dispatch empty or incoming SAVs among subdivided service zones.
Left: Agent-based simulation (FleetPy) of off-peak SoD operations on Munich bus lines. Right: Deep Reinforcement Learning (PPO) architecture for dynamic zonal vehicle dispatching.
Real-World Case Studies
The framework has been evaluated in empirical case studies in both North America and Europe:
Munich, Germany (MVG & TUM): Simulations across 10 municipal bus routes in Munich demonstrated that during off-peak periods, transitioning underperforming fixed lines to autonomous SoD feeders satisfied up to 80% more passenger trips and cut generalized travel costs by up to 50%. Transition studies also revealed operational constraints when managing mixed fleets of human-driven buses and autonomous shuttles.
Dynamic RL Dispatching Gains: When tested against traditional fixed-route services on a real-world bus route in Munich, deep RL-controlled SoD served 16% more passengers (at 13% higher generalized costs), with dynamic PPO zonal control capturing 2.4% more passengers at only 1.4% higher costs compared to static semi-on-demand operations.
Chicago Metropolitan Area (CTA / Pace): Application to suburban-to-urban transit corridors connecting to CTA rail stations demonstrated that hybrid SoD routes significantly expand the spatial reach of trunk rail transit without requiring dedicated park-and-ride infrastructure.
References
2026
Semi-on-demand hybrid transit route design with shared autonomous mobility services
Shared Autonomous Vehicles (SAVs) enable transit agencies to design more agile and responsive services at lower operating costs. This study designs and evaluates a semi-on-demand hybrid route directional service in the public transit network, offering on-demand flexible route service in lowdensity areas and fixed route service in higher-density areas. We develop analytically tractable cost expressions that capture access, waiting, and riding costs for users, and distance-based operating and time-based vehicle costs for operators. Two formulations are presented for strategic and tactical decisions in flexible route portion, fleet size, headway, and vehicle size optimization, enabling the determination of route types between fixed, hybrid, and flexible routes based on demand, cost, and operational parameters. Analytical results demonstrate that the lower operating costs of SAVs favor more flexible route services. The practical applications and benefits of semi-on-demand feeders are presented with numerical examples and a large-scale case study in the Chicago metropolitan area, USA. Findings reveal scenarios in which flexible route portions serving passengers located further away reduce total costs, particularly user costs, whereas higher demand densities favor more traditional line-based operations. Current cost forecasts suggest smaller vehicles with fully flexible routes are optimal, but operating constraints or higher operating costs would favor larger vehicles with hybrid routes. The study provides an analytical tool to design SAVs as directional services and transit feeders, and tractable continuous approximation formulations for planning and research in transit network design.
@article{ng_semi--demand_2024,title={Semi-on-demand hybrid transit route design with shared autonomous mobility services},volume={185},issn={0968090X},doi={10.1016/j.trc.2026.105578},language={en},urldate={2026-02-20},journal={Transportation Research Part C: Emerging Technologies},author={Ng, Max T. M. and Dandl, Florian and Mahmassani, Hani S. and Bogenberger, Klaus},month=apr,year={2026},pages={105578},}
Semi-on-demand off-peak transit services with shared autonomous vehicles — service planning, simulation, and analysis in Munich, Germany
Shared Autonomous Vehicles (SAVs) enable transit agencies to provide more agile and responsive services. This study investigates the implementation of semi-on-demand (SoD) hybrid-route services using SAVs on existing transit lines at off-peak hours, leveraging the additional capacity brought by lower operating costs of SAVs. SoD services combine the cost efficiency of fixed-route buses with the flexibility of on-demand services. SAVs first serve all scheduled fixed-route stops, then drop off and pick up passengers in the pre-determined flexible-route portion, and return to the fixed route. This study addresses three key questions: optimal fleet and vehicle sizes for peak-hour fixed-route services with SAVs, optimal off-peak SoD service planning, and service feasibility during a gradual transition from drivers to autonomous vehicles. The methodology combines analytical modeling for service planning with agent-based simulation for operational analysis. We examine ten bus routes in Munich, Germany and demonstrate that the lower operating costs of SAVs improve service quality through increased frequency and smaller vehicles, even in transition scenarios. At off-peak hours, SoD service satisfies up to 80% more requests and reduces generalized costs by up to 50% compared to traditional fixed routes. However, during transition, maintaining a mixed fleet heavily limits headway improvements, thereby limiting the benefits to short flexible-route lengths.
@article{ng_semi-on-demand_2025,title={Semi-on-demand off-peak transit services with shared autonomous vehicles — service planning, simulation, and analysis in {Munich}, {Germany}},volume={192},issn={0968-090X},doi={10.1016/j.trc.2026.105852},urldate={2026-07-26},journal={Transportation Research Part C: Emerging Technologies},author={Ng, Max T. M. and Engelhardt, Roman and Dandl, Florian and Volakakis, Vasileios and Mahmassani, Hani S. and Bogenberger, Klaus},month=nov,year={2026},keywords={Agent-based simulation, Flexible route, Semi-on-demand, Shared autonomous mobility service, Transit feeder},pages={105852},}
2024
Semi-on-Demand Transit Feeders with Shared Autonomous Vehicles and Reinforcement-Learning-Based Zonal Dispatching Control
This paper develops a semi-on-demand transit feeder service using shared autonomous vehicles (SAVs) and zonal dispatching control based on reinforcement learning (RL). This service combines the cost-effectiveness of fixed-route transit with the adaptability of demand-responsive transport to improve accessibility in lower-density areas. Departing from the terminus, SAVs first make scheduled fixed stops, then offer on-demand pick-ups and drop-offs in a pre-determined flexibleroute area. Our deep RL model dynamically assigns vehicles to subdivided flexible-route zones in response to real-time demand fluctuations and operations, using a policy gradient algorithm — Proximal Policy Optimization. The methodology is demonstrated through agent-based simulations on a realworld bus route in Munich, Germany. Results show that after efficient training of the RL model, the semi-on-demand service with dynamic zonal control serves 16% more passengers at 13% higher generalized costs on average compared to traditional fixed-route service. The efficiency gain brought by RL control brings 2.4% more passengers at 1.4% higher costs. This study not only showcases the potential of integrating SAV feeders and machine learning techniques into public transit, but also sets the groundwork for further innovations in addressing firstmile-last-mile problems in multimodal transit systems.
@inproceedings{ng_rl_itsc_2024,address={Edmonton, AB, Canada},title={Semi-on-{Demand} {Transit} {Feeders} with {Shared} {Autonomous} {Vehicles} and {Reinforcement}-{Learning}-{Based} {Zonal} {Dispatching} {Control}},isbn={979-8-3315-0592-9},doi={10.1109/ITSC58415.2024.10920214},language={en},urldate={2025-03-23},booktitle={2024 {IEEE} 27th {International} {Conference} on {Intelligent} {Transportation} {Systems} ({ITSC})},publisher={IEEE},author={Ng, Max T.M. and Engelhardt, Roman and Dandl, Florian and Mahmassani, Hani S. and Bogenberger, Klaus},month=sep,year={2024},pages={2863--2868},}
2023
Autonomous Minibus Service With Semi-on-Demand Routes in Grid Networks
This paper investigates the potential of autonomous minibuses which take on-demand directional routes for pick-up and drop-off in a grid network of wider area with low density, followed by fixed routes in areas with greater demand. Mathematical formulation for generalized costs demonstrates its benefits, with indicators proposed to select existing bus routes for conversion with the options of zonal express and parallel routes. Simulations on modeled scenarios and case studies with bus routes in Chicago show reductions in both passenger costs and generalized costs compared with existing fixed-route bus services between suburban areas and the central business district.
@article{ng_autonomous_2023,title={Autonomous {Minibus} {Service} {With} {Semi}-on-{Demand} {Routes} in {Grid} {Networks}},volume={2677},issn={0361-1981},doi={10.1177/03611981221098660},language={en},number={1},urldate={2022-12-26},journal={Transportation Research Record},author={Ng, Max T. M. and Mahmassani, Hani S.},year={2023},pages={178--200},}