# Learning to Solve Resource-Constrained Project Scheduling with Duration Uncertainty using GNNs Source: https://www.youtube.com/watch?v=MTod4inkPX4 Recap page: https://rapidrecap.app/video/MTod4inkPX4 Generated: 2025-11-22T01:04:59.426+00:00 --- ## Quick Overview The Wheatley AI scheduling system successfully solves resource-constrained project scheduling problems under duration uncertainty by using a novel approach that combines Deep Reinforcement Learning (DRL) with a novel GNN-based approach, significantly outperforming traditional heuristics like Shortest Processing Time (SPT) and CP50 by achieving a notably lower expected makespan across various test cases, including a large-scale, real-world industrial dataset. **Key Points:** - Wheatley, an AI scheduling method combining DRL and GNNs, addresses Resource-Constrained Project Scheduling with Duration Uncertainty (RCPSP-DU). - Wheatley outperformed traditional heuristics like Shortest Processing Time (SPT) and CP50, achieving a 1.3% lower expected makespan compared to the best solutions found by CP50. - The core innovation involves using a novel GNN structure to encode complex, multi-perspective information (precedence, resource constraints, probability distributions) into resource nodes. - The DRL agent learns a Priority Dispatch Rule (PDR) that intelligently selects the next task based on future uncertainty and available resources, avoiding the pitfalls of simple rules. - In a large-scale test case involving 120 jobs from a public industrial dataset, Wheatley successfully found robust, feasible schedules that traditional solvers failed to solve within the 60-second time limit. - Wheatley embeds resilience directly into the scheduling decision, allowing it to anticipate and mitigate potential future resource conflicts before they occur. ![Screenshot at 00:48: A speaker introduces the challenge of defining the core difficulty in RCPSP-DU, noting that while it sounds simple, the reality of uncertainty makes it deceptively complex.](https://ss.rapidrecap.app/screens/MTod4inkPX4/00-00-48.png) **Context:** The video discusses a new Artificial Intelligence approach, named Wheatley, designed to tackle the complex problem of Resource-Constrained Project Scheduling with Duration Uncertainty (RCPSP-DU). This problem is common in sectors like aerospace manufacturing and logistics where project schedules must account for unknown task durations while respecting resource limits. Traditional methods often rely on reactive scheduling or simple heuristics, which perform poorly when uncertainty is high, leading to schedule breakdowns and operational nightmares. ## Detailed Analysis The discussion centers on overcoming the Resource-Constrained Project Scheduling with Duration Uncertainty (RCPSP-DU) problem, which plagues industries like aerospace manufacturing and logistics. The core issue is that simple, reactive scheduling methods fail when uncertainty (like variable task durations) causes cascading delays. The proposed solution, Wheatley, leverages Deep Reinforcement Learning (DRL) combined with a specialized Graph Neural Network (GNN) architecture. The GNN encodes complex information—precedence, resource constraints, and uncertainty—into resource nodes. The DRL agent then learns an optimal Priority Dispatch Rule (PDR) that considers both precedence and resource availability over time, effectively embedding resilience into the scheduling process. Wheatley was benchmarked against traditional heuristics like Shortest Processing Time (SPT) and CP50. The results were overwhelmingly positive: Wheatley achieved a 1.3% lower expected makespan than the best traditional solutions, demonstrating superior performance, especially when dealing with the inherent uncertainty and the need for robust, repeatable schedules. ### The RCPSP-DU Challenge - Every major industrial sector faces this challenge - Schedules are built on perfect assumptions that reality shatters - Uncertainty immediately complicates scheduling. ### The Wheatley AI Solution - Details an AI framework addressing duration uncertainty - Uses advanced Deep Reinforcement Learning (DRL) - Employs a GNN to model the complex reality of tasks, resources, and uncertainty. ### GNN Architecture and Learning - The DRL agent learns an optimal Priority Dispatch Rule (PDR) from scratch - It uses a GNN to encode precedence, resource constraints, and uncertainty information into resource nodes. ### Performance Metrics - Wheatley outperformed traditional heuristics (SPT, CP50) - Achieved 1.3% lower expected makespan than the best CP50 solution - Showed superior generalization ability across different problem types and scales. ### Real-World Application - Wheatley successfully scheduled a 120-job industrial dataset where traditional solvers failed within the time limit - The agent learns to anticipate future resource conflicts by looking forward across multiple potential outcomes. ![Screenshot at 00:00: The video opens with an invitation to become a member, set against a graphic representing audio waves.](https://ss.rapidrecap.app/screens/MTod4inkPX4/00-00-00.png) ![Screenshot at 00:14: The speaker introduces the scope of the problem, mentioning that it affects major industrial sectors from aerospace to construction.](https://ss.rapidrecap.app/screens/MTod4inkPX4/00-00-14.png) ![Screenshot at 00:39: The speaker details the core paper, which introduces Wheatley, an AI framework designed to address duration uncertainty in scheduling.](https://ss.rapidrecap.app/screens/MTod4inkPX4/00-00-39.png) ![Screenshot at 01:55: A visual representation of a fluctuating waveform illustrates the uncertainty inherent in project duration estimates.](https://ss.rapidrecap.app/screens/MTod4inkPX4/00-01-55.png) ![Screenshot at 03:33: The speaker contrasts the failed deterministic approach with the successful probabilistic framework employed by Wheatley.](https://ss.rapidrecap.app/screens/MTod4inkPX4/00-03-33.png)