Vehicle Scheduling Algorithms

After thinking deeply and practically as an operations research practitioner working on transforming Toyota North America’s supply chains and logistics, here’s how I’ve come to think about yard scheduling and sequencing in real-world operations. In yard logistics, deciding how to handle scheduling and sequencing often comes down to a fundamental design choice: do you build one end-to-end optimization model, or break the problem into a hierarchical structure? Both approaches have merit, but the right choice depends on the maturity of your operation, your data quality, and your need for responsiveness. An end-to-end model treats the entire yard as a single optimization problem. From the moment a vehicle arrives at the gate to the point it exits after fueling, staging, and processing, every step is scheduled together. The advantage is clear: this global view can yield truly optimized flows, minimize bottlenecks, and align all decisions toward throughput and efficiency. However, this approach is heavily reliant on high-fidelity data across all zones of the yard. If there are unexpected delays (like a worker calling in sick, a fueling station going offline, or a shuttle arriving late) the whole plan can become fragile. In practice, this kind of system also tends to be computationally heavier and slower to react in real time unless reoptimization is thoughtfully designed. A hierarchical model breaks the yard down into zones or functions. High-level scheduling might determine when a shuttle should move workers, or when a batch of cars should be released to fueling. Then, each zone locally decides how to sequence its tasks. This makes the system more resilient, local disruptions can be handled without having to re-optimize everything. It’s often easier to implement and more forgiving of messy, real-world variability. But the downside is that decisions made independently in one zone may not be optimal when viewed from the yard’s perspective as a whole. You may reduce wait time in one area only to create backups downstream. In my experience working in yard scheduling, the best path forward is staged. If your data systems are still maturing, start with a hierarchical model. It’s practical, easier to maintain, and more robust to the day-to-day volatility of yard operations. As your visibility improves and your systems become more integrated, evolve toward an end-to-end approach, or better yet, implement a hybrid system. In this setup, global optimization runs periodically throughout the day, reoptimizing as new information comes in (whether it’s a backlog at a station, a no-show worker, or a delay in vehicle arrivals). Meanwhile, local zones retain autonomy to adjust sequencing based on real-time conditions. This balance between strategic coordination and operational flexibility is where the real performance gains are unlocked. #YardScheduling #Optimization #SupplyChainExecution #OperationsResearch #DecisionIntelligence #RealTimeOptimization #Logistics

See this post "Inside Glovo’s deliveries optimization" from Glovo. Efficiency is the key to success in the instant-delivery business. The Matching Team from Glovo has developed innovative methods to optimize order assignments for couriers. Here’s how they do it: → Assignment Problem: Glovo tackles the challenge of assigning the best courier for each order by using algorithms like the Hungarian Algorithm and Minimum Cost Flow. They aim to reduce courier waiting times and improve delivery speed without violating business constraints. → Bundling Strategy: Combining multiple orders into a single delivery, or bundling, allows them to increase efficiency, but it’s not simple! They consider real-time factors like preparation time, travel distance, and traffic conditions. → Multibundling & Scaling: They extended bundling to multiple orders across different locations, but handling such complexity required solving the Vehicle Routing Problem (VRP) efficiently, especially when considering millions of orders and couriers. → Architecture: To manage ~20 million assignments monthly (that's back in 2023, they rely on an advanced system architecture using relational databases, Redis for geo-indexing, and scalable instances. All decisions are processed in under 10 seconds per city. → Continuous Optimization: Real-time data and predictions help them adapt to changing circumstances, ensuring we always make the most efficient choices. With technology and algorithms, they're pushing the boundaries of what’s possible in delivery optimization. 🚴♂️📦

In the context of electrifying paratransit, we investigate heuristic scheduling of minibus taxis in South Africa's eventual electrified paratransit. Link: https://lnkd.in/dCYvDaMd The proposed method supports: - multiple ranks/depots, - mixed fleet (electric and ICE) deployment, - number of charging point optimisation, - partial and opportunistic charging of vehicles, and - scheduling with load shedding (or to avoid peak tariff times) Abstract: The predominant mode of public transport in South Africa originates from the informal sector, specifically “paratransit”. Vehicles carry up to 23 passengers and are still propelled by internal combustion engines. We investigate the feasibility of using electric vehicles without negating the loss of opportunities by drivers and owners. We propose that scheduling of the electric vehicles is one important cornerstone towards electrification. We developed a fast-executing heuristic scheduling algorithm that allows for multiple vehicle depots in the transport network; simultaneous electric and internal combustion engine vehicle deployment; determining the number of charging stations; partial charging; and scheduled charging with intermittent electricity supply. The scheduling algorithm achieves the minimum number of vehicles to execute the passenger demand in shorter total distances, outperforming current approaches. The algorithm demonstrated multi-objective optimisation by minimising the vehicles, the number of charging stations, and the average trip delays of a schedule. With Jacques Wust and James Bekker. Industrial Engineering | Stellenbosch University Electrical & Electronic Engineering Department (Stellenbosch University) Related paper: Scheduling with mixed fleets to improve the feasibility of electric minibus taxis: a case scenario of South Africa: https://lnkd.in/deBrfug8 For more: https://lnkd.in/dukgNDkj

Vehicle route optimisation for NEMT, prioritising high-needs passengers! For scheduling problems in non-emergency medical transportation (NEMT), non-emergency patient transportation (NEPT) and community transportation, passengers need to be picked up or dropped off at specific times and can’t be onboard too long. In real life it’s common that not all constraints can be met, i.e. some passengers must be served late or have a journey time exceeding the maximum allowed (because other passengers are served on the way). For some passengers, this is OK, however for the elderly, or those requiring greater care, this may not be acceptable. Our ODL Live scheduling engine can set different priorities for different passengers. In this example, not all passengers can be served on time. We place passengers into 3 priorities based on their needs and use custom late-time and on-board time penalty functions in the optimisation model to ensure higher priority passengers get the best service. We set higher priority passengers to have more expensive penalty functions, so the optimiser tries harder to reduce these compared to the lower cost functions. The video shows the optimiser running for this problem. As expected, higher needs passengers get the better service (not late, not on-board too long). #routeoptimization #NEMT #algorithms #passengertransportation

🚀 New Project for Resume/CV — Machine Learning Engineer | AI Engineer Real-time Delivery System (Solving the Vehicle Routing Problem) with Genetic Algorithms I recently worked on an optimization project where I applied Genetic Algorithms (GA) to tackle the Vehicle Routing Problem (VRP) — a core challenge in logistics, food delivery, production units, and wholesale/retail distribution systems. 🔑 Highlights: Implemented GA using DEAP to optimize vehicle routes. Designed a fitness function for efficient route evaluation. Experimented with GA on a maze solver before extending to VRP. Visualized solutions with Matplotlib, showing improvements over generations. 💡 Key Takeaway: This project enhanced my understanding of evolutionary algorithms and their real-world applications in logistics optimization. 📂 GitHub: https://lnkd.in/d7b96dBW 📝 Medium Blog: https://lnkd.in/d3p8GhqV #geneticalgorithms #optimization #vehicleroutingproblem #logistics #supplychain #machinelearning #mlproject #aiproject #evolutionarycomputing #python #datascience #operationsresearch #mlengineer #aiengineer #resume #cv