.png){kind=icon}
.png){kind=icon}
Return
RL-based anticipatory routing and matching in on-demand ridepooling
.png){kind=icon}
OA
AI
X
A
DOI：10.1016/j.trc.2026.105898.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Demonstrate that fixed anticipation parameters cannot adapt to evolving supply-demand conditions, limiting system efficiency over time. • Propose a GNN+TD3 reinforcement learning framework that dynamically adjusts the level of anticipation in real-time routing and matching decisions. • Discover a “seed-and-harvest” behavioral pattern emerging from RL, where the policy invests in future fleet positioning early and switches to greedy optimization as the system saturates. • The policy generalizes robustly across fleet sizes (1,000–2,000 vehicles) and stochastic travel times without retraining.
Keywords:
On-demand ridepooling
Reinforcement learning
Anticipatory matching
Anticipatory routing
Graph neural networks
Journal
IF:
7.9
Papers:
4.7K
Citations:
3.2W
Organization
No organization information available
Cited Papers