Dynamic multi-agent deep reinforcement learning-based pricing and incentivization approach in multimodal transportation networks

Researchers propose a multi-agent deep reinforcement learning framework to optimize pricing and incentives across shared mobility services and public transport, balancing competing objectives between authorities, providers, and commuters. Simulations demonstrate the approach reduces congestion by 20%, lowers emissions by 10%, and doubles public transport profit while improving equity.

This research addresses a fundamental challenge in modern urban mobility: coordinating conflicting stakeholder interests through algorithmic decision-making. Transportation systems involve three parties with opposing goals—authorities prioritizing sustainability and equity, SMS providers maximizing revenue, and travelers minimizing personal costs. Traditional optimization approaches struggle with such multi-objective problems at scale. The proposed framework uses dual reinforcement learning agents that learn adaptive strategies through continuous interaction with the transportation network, adjusting to real-time demand fluctuations and congestion patterns.

The work builds on growing adoption of machine learning in transportation planning, where cities increasingly leverage data-driven approaches to optimize service delivery. Dynamic pricing has proven effective in ride-sharing platforms, while incentive-based demand management shows promise in public transit adoption. This research bridges both domains, treating them as interdependent rather than competing systems.

For urban planners and policymakers, the framework offers significant value by demonstrating that conflicting objectives need not be zero-sum. The 20% cost reduction for commuters alongside nearly doubled public transport revenue suggests efficiency gains benefit multiple stakeholders. However, real-world implementation faces practical challenges: regulatory frameworks may resist dynamic pricing, equity concerns arise when incentives disproportionately benefit affluent areas, and user adoption depends on transparency in algorithmic decision-making.

The research validates through simulation over a three-hour peak period, but scaling to full-day operations and diverse urban contexts requires validation. Future applications will likely focus on integrating real-time data feeds and testing in pilot cities, particularly those with emerging mobility ecosystems where algorithmic coordination can take root before entrenched practices calcify.

• →Multi-agent reinforcement learning can simultaneously optimize revenue for SMS providers and sustainability goals for public authorities without sacrificing either objective.
• →Dynamic incentivization reduced peak congestion while lowering commuter costs by approximately 20% in simulation experiments.
• →The framework demonstrates 10% emissions reduction through rebalancing modal choice between shared mobility and public transport.
• →Real-world deployment requires addressing regulatory concerns around dynamic pricing transparency and equitable access across neighborhoods.
• →The research provides a template for AI-driven coordination in systems with structurally opposing stakeholder incentives.