How agentic AI solves complex scheduling problems
summary
Simulated an airline with 7,500 flights over 30 days, 80 aircraft, 900 crew and 25 airports and compares MILP (4%), genetic (4.8%) and simulated annealing (6.8%) to reinforcement learning. Single-agent PPO reaches 68% coverage; a multi-agent MAPPO setup with shared early network layers, adaptive Lagrange penalties and per-agent heads yields MAPPO (multi-agent PPO) and scheduled 87% of 7,500 flights, with 92% operational efficiency and uniform aircraft/crew utilization.
AI-written summary · about
chapters (11)
- 0:00Introduction to optimization
- 1:24Airline scheduling challenges
- 5:53Limitations of current solutions
- 8:37Defining agentic AI
- 10:36Reinforcement learning basics
- 15:23Transition to multi-agent RL
- 17:32MARL system architecture
- 22:21Performance results
- 24:12Design and implementation keys
- 26:44Conclusion and Q&A
- 30:30Technical deep dive Q&A
related talks
11mUsing Deep Reinforcement Learning to Coordinate Multi-Modal Journey Planning with Limited ...
22mTaking Flight with AI and ML: Revolutionizing Airline Ops
37mMultiagent Reinforcement Learning: Rollout and Policy Iteration
19mThe Rise of Agentic AI and Its Transformative Potential
31mTaming Turbulence with Code: Developing AI-powered Solutions for Airline Disruptions
23mAlphaDow: Reinforcement Learning for Industrial Production Scheduling
45mLLMs can't optimize schedules, but AI can!
26mAI Agents as Financial Co-Pilots | AI-Driven Strategy & Decision Making