Smart Routing: Real-Time Transport Optimization with Fabric

Description

Discover how Microsoft Fabric enables real-time transportation optimization. Learn to ingest GPS data with Eventstream, visualize routes on live maps, and integrate linear optimization models to minimize cost and emissions. Explore how data agents and AI Foundry can help in identifying EV-ready routes based on charging infrastructure.

Key Takeaways

• Truck dispatching problem, travelling salesman problem …
• NP-hard problem (Non-deterministic Polynomial-time hard)
• Common constraints VRP must solve for:
• →Provides optimized sequence for package deliveries.
• Handles parcel deliveries to residential areas
• 3 delivery trucks: AC-01, AC-02, AC-03
• Approx. 50 deliveries daily

My Notes

Action Items

• [ ]

Resources & Links

Slides

📥 Download Slides

Smart Routing: Real-Time Transport Optimization with Fabric
Solving the Vehicle Routing Problem with PyVRP on Microsoft Fabric
Agenda

1. Meet the speakers
2. The Problem
3. Tech Stack
4. Solutioning & Demo
   Meet the Speakers
   Emmanuel Huygens
   Robert Leal
   Cloud & Data Engineer
   Data & AI Engineer
   The Vehicle Routing Problem (VRP)
   • Truck dispatching problem, travelling salesman problem …
   → "What is the optimal set of routes for a fleet of vehicles
   to traverse to deliver to a given set of customers?“
   • NP-hard problem (Non-deterministic Polynomial-time hard)
   → Gets exponentially more challenging with each new variable
   e.g. more vehicles, more stops, more constraints
   → Requires heuristic approach
   • Common constraints VRP must solve for:
   → Vehicle capacity, delivery time windows, load balancing,
   pickup and delivery pairing, Q depots …
   VRP Infamy
   ORION On-Road Integrated Optimization and Navigation (2016)
   Last Mile Routing Research Challenge (2021)
   →Estimated cost of 250m USD
   →Competition to train models on delivery routes
   →Provides optimized sequence for package deliveries.
   →Bridge algorithmic optimization with real-world
   driver knowledge
   →Approx. 55k drivers rely on it
   →220 Participating academic teams
   →Projected annual savings exceed 300m USD
   →Drivers frequently deviate from computed routes
   due to tacit knowledge of traffic patterns, parking,
   etc..
   To help solve this problem, we need data..
   With data, we can use:
   ▪ Microsoft Fabric as the unified data platform
   ▪ Fabric Maps as the visual layer
   ▪ Azure Maps API and PyVRP for route optimization
   ▪ Eventhub and Eventstreams for real-time data ingestion
   ▪ IoT Datasources for fleet telemetry (simulated using FunctionApp)
   ▪ Real-Time Dashboards and Live Maps
   ▪ Fabric Data Agents for interactive fleet feedback
   ▪ Bonus: Model Planetary Computer in Fabric Maps
   Case Study: Beltway Couriers
   • Logistics firm
   • Handles parcel deliveries to residential areas
   • Greater Houston Area
   • 3 delivery trucks: AC-01, AC-02, AC-03
   • Approx. 50 deliveries daily
   • Morning & Afternoon shifts
   Beltway Couriers: Logistics Manager
   Daily orders
   Order insights
   ETL Medallion
   Beltway Couriers: Logistics Manager (Demo)
   Beltway Couriers: Logistics Manager (Demo)
   Beltway Couriers: Logistics Manager (Demo)
   Beltway Couriers: Logistics Manager (Demo)
   Beltway Couriers: Logistics Manager (Demo)
   GeoJSON:
   • Fabric Notebooks
   • Azure Maps API
   • Python libraries
   How PyVRP Finds Solutions
   Soft Constraints
   Time windows and capacity treated as soft constraints
   with auto-adjusted penalties during local search
   Target Feasibility
   ~43% of local search runs produce feasible solutions
   — balances exploration vs. exploitation
   Our FleetPulse runs: 30,000+ iterations in ~30 seconds, converging on near-optimal solutions for 44 stops across 3 vehicles.
   The Constraints
   Morning Shift – 20 Orders Solved
   PLANNING NOTEBOOK
   Afternoon Shift – 24 Orders Solved
   PLANNING NOTEBOOK
   When Reality Hits – Failure Processing
   Result: 1 order rescheduled into the afternoon pool. 4 deferred to next day. Re-optimization triggered automatically.
   The Solver Adapts in 30 Seconds
   RE-OPTIMIZATION
   What “Optimal” Actually Means
   Single Cost Function
   Distance + duration + overtime penalties + load
   balance — combined into one score. Lower =
   better.
   Provably Better
   Before/after cost comparison proves the solver
   beats naive approaches like "append to nearest
   truck."
   Complete Operational Plan
   Ordered sequences, arrival/departure times,
   overtime warnings, road-following polylines.
   Dispatch Integration
   routes.json → dispatch system → drivers see
   updated sequences instantly.
   Why This Matters on Fabric
   Traditional approach: 5+ separate services to provision, secure, monitor, and pay for. Fabric: one platform.
   Beltway Couriers: Logistics Manager (Demo)
   Beltway Couriers: Logistics Manager (Demo)
   Microsoft Building Footprints
   Machine learning detected polygons
   Biodiversity intactness
   Global terrestrial biodiversity intactness
   Stage 1: Semantic Segmentation
   Recognizing building pixels using deep neural networks
   Biomass carbon density
   HGB: Harmonized Global Biomass
   Digital Elevation Models (30m)
   ALOS: Advanced Land Observing Satelite
   Stage 2: Polygonization
   Microsoft Building Footprints
   Converting building pixel detections into polygons
   Machine learning detected polygons
   How was
   the session?
   Complete Session Surveys in
   for your chance to WIN
   PRIZES!