Last‑Mile: Hotspot‑Based Path‑Sharing (IEEE T‑ITS)
Multi‑objective (cost & satisfaction) • −16% distance • −15% fleet • +50% utilization
This page documents a multi‑year research program led by Vaskar Raychoudhury and collaborators, progressing from decentralized taxi ride sharing to reinforcement‑learning‑based delivery logistics. It includes a curated timeline, research themes, collaborator network, and full publication details.
Our work tackles the core challenges in dynamic, distributed mobility coordination. For taxis, we designed decentralized algorithms for matching and routing using localized communication and metaheuristics (PSO/ACO). We introduced causal congestion graphs and hotspot recommendation to embed real‑time traffic and demand. The program evolved into last‑mile delivery, where we formulated path‑sharing for multi‑drop deliveries and built DeliverAI, a reinforcement‑learning multi‑agent system that balances consumer satisfaction and operating cost while improving fleet utilization and reducing miles.
Multi‑objective (cost & satisfaction) • −16% distance • −15% fleet • +50% utilization
RL‑based distributed path‑sharing • −13% distance • +12% fleet reduction • +50% utilization
Ant Colony Optimization • Multi‑objective Pareto optimization • Up to 79.65% success • Chicago data
Publisher–subscriber model • Particle Swarm Optimization • 91.74% non‑peak success
Ant Colony Optimization • Multi‑objective • Up to 77% shared‑ride acceptance
Asynchronous passenger–taxi protocol • 76% share success • 97.5% peak occupancy • Chicago data
Centralized • Distributed • Hybrid • Open challenges
Distributed, localized congestion detection • 65% prediction accuracy • Shanghai taxis
Distributed coordination (TSA) • Hotspot Recommendation (HRA) • SFO & NYC datasets
TRS algorithm • 33% higher ride share • Shanghai GPS traces
Decentralized taxi ride matching (TRS), taxi–passenger selection (TSA) and hotspot modeling (HRA), and causal congestion graphs for traffic‑aware dispatch.
First comprehensive survey of ride‑sharing architectures—centralized, distributed, and hybrid—mapping open problems and design trade‑offs.
Localized asynchronous communication for dynamic topologies; route optimization via PSO and ACO to boost share success, occupancy, and reduce miles.
Proposed cooperative and adaptive distributed ride‑sharing system using Ant Colony Optimization and Pareto‑optimal multi‑objective formulation, showing up to 79.65% success in large‑scale Chicago taxi data.
Recognized structural parallels between ride sharing and multi‑drop deliveries. Introduced DeliverAI—an RL multi‑agent system for path‑sharing under time and service constraints. Extended with hotspot‑based routing and explicit cost–satisfaction trade‑offs, cutting fleet size and VMT while maintaining delivery SLAs.