Traditional transit optimization treats passengers like uniform blocks of cargo. The goal is straightforward: minimize average travel time, minimize fuel consumption, or maximize vehicle throughput. However, humans are not cargo. We experience stress, fatigue, and frustration. An optimization algorithm that reduces average wait time by 30 seconds at the cost of leaving a handful of passengers stranded for an hour is technically efficient but practically broken.
As Jiafu Wan et al. note in Cloud-assisted Industrial Systems and Applications, ignoring the human experience creates systems that are socially and practically unsustainable. To build truly robust transit networks, modern modeling must shift from pure efficiency to integrating psychological comfort and subjective thresholds directly into core optimization loops.
The Non-Linearity of Passenger Disutility
Why does optimizing for the "average" fail? Because human frustration does not scale linearly with time or crowding. Waiting 20 minutes for a bus feels significantly worse than twice a 10-minute wait.
In Bus Frequency Optimization: When Waiting Time Matters in User Satisfaction, SONGSONG MO et al. demonstrate that passenger satisfaction degrades non-linearly as wait times increase. Instead of minimizing average wait times, the authors propose a threshold-based approach that maximizes the number of passengers served within an acceptable waiting tolerance. This principle underpins the FASTS scheduling assistant developed by SONGSONG MO et al. in FASTS: A Satisfaction-Boosting Bus Scheduling Assistant, which prioritizes meeting a standard of acceptable service for the majority over marginal average improvements.
A similar non-linear threshold governs onboard crowding. Passengers do not evaluate crowding in a vacuum; their tolerance depends heavily on how long they will be on the vehicle. In Analysis of bus passenger comfort perception based on passenger load factor and in-vehicle time, Xianghao Shen et al. quantified this coupling, showing that maintaining acceptable comfort over a 40-minute journey requires keeping the passenger load factor strictly below 0.36.
Quantifying Comfort via Revealed Preferences
How do we capture these subjective psychological constraints mathematically? We can observe them through the choices passengers make when confronted with discomfort.
During peak hours, commuters frequently exhibit the "left-behind" phenomenon. As modeled by Juanjuan Zhao 0001 et al. in Estimation of Passenger Route Choice Pattern Using Smart Card Data for Complex Metro Systems, passengers will voluntarily skip the first available train to wait for a less crowded one where they have a higher chance of finding a seat.
Standard Routing: [Origin] ---> (Shortest Time / Crowded Train) ---> [Destination]
Comfort Routing: [Origin] ---> (Longer Route / Guaranteed Seat) ---> [Destination]
^ Commuters pay a "time tax" for comfort
This behavior reveals a quantifiable trade-off. By analyzing smart card data, Alejandro Tirachini et al. in Valuation of sitting and standing in metro trains using revealed preferences proved that commuters are willing to deliberately extend their travel times by taking longer routes just to secure a seat.
Leveraging this insight, Danhuai Guo et al. developed personalized recommendation services in How to Find a Comfortable Bus Route - Towards Personalized Information Recommendation Services that prioritize lower passenger density over absolute speed. This aligns with findings by Long Cheng et al. in Improving Travel Quality of Low-Income Commuters in China, which show that latent psychological attitudes toward comfort and reliability are more significant drivers of transportation mode choice than traditional objective variables like time and cost.
The Efficiency-Comfort Trade-off in Automated Control
When automated systems or reinforcement learning (RL) agents manage transit infrastructure, optimizing purely for mathematical throughput often results in erratic behaviors that alienate human users.
In traffic signal control, Arthur Müller demonstrates in Safe and Psychologically Pleasant Traffic Signal Control with Reinforcement Learning using Action Masking that while unconstrained RL agents maximize vehicle flow, they do so by executing rapid, unpredictable signal changes that stress drivers. Introducing constraints for "psychological pleasantness" and predictability causes a measurable decrease in raw mathematical efficiency, but it is necessary to ensure public trust.
A similar "comfort tax" appears in shared mobility networks. In Community-Based Trip Sharing for Urban Commuting, Mohd. Hafiz Hasan et al. showed that enforcing psychological preferences—such as maintaining stable driver-passenger pairings—drastically diminishes the environmental benefits of ride-sharing compared to mathematically optimal, flexible daily re-assignments.
Furthermore, aggressive operational interventions used to combat bus bunching, such as stop-skipping, often trigger severe passenger dissatisfaction and perceived inequity. As reviewed by Ying Yang et al. in An overview of solutions to the bus bunching problem in urban bus systems, transit models must incorporate passenger psychology to balance operational efficiency with social acceptance. This is echoed by Liping Ge et al. in Robustness and disturbances in public transport, who argue that public transport robustness models must move away from purely operator-side costs to incorporate psychosocial factors like traveler stress.
Redefining Travel Time as an Experience
If travel time is treated strictly as a cost to be minimized, transit design becomes a race to the bottom. However, when systems optimize for the subjective experience, the perceived burden of time shifts.
Cristian Consonni et al. introduced the MoTiV dataset in What's Your Value of Travel Time? Collecting Traveler-Centered Mobility Data via Crowdsourcing to capture the "worthwhileness" of a trip, treating travel time as an opportunity for productivity, fitness, or leisure rather than a mere waste.
When the quality of the experience is elevated, human tolerance for traditional friction points changes:
- On-Demand Flexibility: Lotte Notelaers et al. found in Preferences for stop-based pooled automated vehicles in mode choice models across trip purposes in Flanders, Belgium that passengers demonstrate a higher tolerance for walking and waiting when using on-demand pooled automated services, because the perceived quality of the system mitigates the psychological disutility of waiting.
- Proportional Cost: In itinerary recommendation, Sajal Halder et al. showed in Efficient itinerary recommendation via personalized POI selection and pruning that user satisfaction improves when the system ensures that active engagement time at a destination proportionally justifies the associated queuing and travel costs.
- Fairness Integration: Frameworks like HCRide, developed by Lin Jiang 0007 et al. in HCRide: Harmonizing Passenger Fairness and Driver Preference for Human-Centered Ride-Hailing, demonstrate that multi-objective reinforcement learning can successfully harmonize platform efficiency with passenger waiting fairness and driver preferences, proving that human-centric constraints do not inherently degrade system performance.
Key Foundational Work
Analysis of bus passenger comfort perception based on passenger load factor and in-vehicle time (2016) — contributed 8 concept elements to this direction.
Cloud-assisted Industrial Systems and Applications (2016) — contributed 3 concept elements to this direction.
Estimation of Passenger Route Choice Pattern Using Smart Card Data for Complex Metro Systems (2016) — contributed 2 concept elements to this direction.