Using Deep Reinforcement Learning to Coordinate Multi-Modal Journey Planning with Limited ...
summary
Multi-agent deep RL is applied to SUMO using PPO via RLlib and Ray to generate coordinated multimodal plans for an event scenario with 1,000 agents and a synthetic background of 20,000 movers. Learned policies beat a naive Google-Maps–style baseline on average travel time and lateness, handle parking requirements and vehicle-ownership constraints, and reveal public transit remains underselected versus bikes/motorbikes; coordination effects depend on reward design, and all code and environments are published on GitHub.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:24Overview
- 1:17Reinforcement Learning
- 1:57The Goal: Just-in-Time Arrival
- 2:33Travelers (Agents) Coordination
- 3:46MADRL Infrastructure
- 4:33In the Paper
- 6:03Learning Environment • 1000 Agents
- 7:46Metrics
- 8:25Results
- 9:43Conclusions
related talks
15mA Comparison of Reinforcement Learning Agents Applied to Traffic Signal Optimization
20mEnergy Optimized Green Light Assist in Varying Traffic Scenarios Using Reinforcement Learning
24mDeepCoMP: Multi-agent Reinforcement Learning for Multi-cell Selection in 5G and Beyond
35mCrowd dynamics, RL and Unity: A Journey
14mConnecting Reinforcement Learning to Simulation Software
19mTestbed for Development of AI Solutions to Boost Operational Efficiency in Transportation Networks
47mHow agentic AI solves complex scheduling problems
37mMultiagent Reinforcement Learning: Rollout and Policy Iteration