Centram Centre for Applied Artificial Intelligence

AI & Machine Learning Course for Transport and Logistics Management

Course description

This course covers AI and ML techniques applied to transportation and logistics problems: demand forecasting, travel-time estimation, routing and scheduling (VRP), fleet and inventory management, predictive maintenance, dynamic pricing, and autonomous/connected mobility. Students learn methods (time series, supervised learning, optimization, RL, graph neural networks), tools, datasets, and deployable workflows through hands-on labs and a final project.

Learning objectives

  1. Understand transportation and logistics business problems and their ML formulations.
  2. Pre-process and analyse spatio-temporal logistics/traffic data.
  3. Apply supervised and unsupervised ML models to forecasting, classification, and anomaly detection.
  4. Formulate and solve routing and scheduling problems using ML/OR and hybrid methods.
  5. Design and evaluate reinforcement learning solutions for dynamic routing and fleet control.
  6. Use industry tools (Python, scikit-learn, TensorFlow/PyTorch, OR-Tools, SUMO) and public datasets.
  7. Appreciate ethical, privacy, safety, and sustainability implications.

Course Outline


Overview & problem framing

  1. Transport & logistics domains, examples (last-mile delivery, urban mobility, freight, ports).
  2. How to frame problems for ML vs optimization vs hybrid solutions.
    Course project kick-off: choose topics/data.

Data for transport & logistics

  1. Spatio-temporal data types: GPS trajectories, trip logs, sensor/IoT, telematics, event logs.
  2. Data cleaning, map-matching, spatial joins, aggregation, feature engineering.

Tools: GeoPandas, PostGIS, Folium, OSMnx.

Time series & demand forecasting

  1. Classical methods (ARIMA, exponential smoothing) and ML-based (XGBoost, LSTM, Prophet).
  2. Evaluation metrics, seasonality, holidays, exogenous variables.
  3. Lab: forecast station/zone demand (bike-share or taxi dataset).

Travel time and ETA estimation

  1. Supervised learning for travel time estimation, feature engineering (road network, time-of-day).
  2. Spatio-temporal models: convolutional, RNNs, graph neural networks (GNNs).
  3. Datasets: METR-LA, PEMS-BAY; tools: PyTorch Geometric.

Freight and inventory forecasting / capacity planning

  1. Demand forecasting for freight, inventory models, lead times.
    Using ML to predict demand spikes and optimize safety stock.
  2. Case study: e-commerce peak season planning.

Route planning & Vehicle Routing Problems (VRP)

  1. Classical VRP, VRP with time windows, pickup & delivery, heterogeneous fleets.
  2. Exact vs heuristics (local search, tabu, genetic algorithms) and OR-tools.
    Lab: solve capacitated VRP with OR-Tools on sample dataset.

Integrating ML with optimization

  1. Predict + optimize pipelines, data-driven constraints, surrogate models.
    Learning to predict cost/edge weights; two-stage and end-to-end approaches.
  2. Example: predicted travel times feeding VRP.

Dynamic routing & real-time decision-making

  1. Online/dynamic VRP, dispatching, re-routing, ride-pooling concepts.
    Use of streaming data, latency considerations, simple RL baselines.
  2. Tools: SUMO for traffic simulation.

Reinforcement learning for routing & fleet management

  1. RL basics, MDPs for transport tasks, deep RL (DQN, PPO), multi-agent RL.
  2. Environment design, reward shaping, simulation/real-world bridging.
  3. Lab: train RL agent for simplified dispatch in simulated environment.

Predictive maintenance & anomaly detection

Telematics data, remaining useful life (RUL) prediction, fault detection.
Unsupervised and supervised techniques; survival analysis basics.
 Case study: fleet maintenance scheduling.

Pricing, demand management & market design

  1. Dynamic pricing for ride-hailing/logistics, surge mechanisms, fairness considerations.
  2. Causal inference basics for pricing experiments and A/B testing.

Autonomous vehicles, connected mobility & sustainability

  1. Platooning, routing for EVs (range constraints, charging scheduling).
    Emissions-aware routing, modal shift, digital twins.
    Ethics, privacy, safety, regulatory considerations.

Project presentations & wrap-up

           Final project demos, evaluation, lessons learned and industry trends.

           Assessments

  1. Weekly labs / programming exercises: 30%
  2. Midterm exam or project milestone: 20%
  3. Final project (report + demo): 35%
  4. Participation / quizzes / reading summaries: 15%

Labs / practical

  1. Data cleaning & map-matching of GPS traces.
  2. Demand forecasting (bike-share or taxi).
  3. Travel time estimation with gradient-boosting + GNN baseline.
  4. Implement CVRP with OR-Tools and compare heuristics.
  5.  RL dispatch simulator on SUMO or custom environment.
  6.  Predictive maintenance using telematics data.

Final project ideas

  1. Real-time dynamic routing for last-mile delivery (simulation + RL or heuristics).
  2. Demand forecasting and capacity planning for a city bike-share system.
  3. Travel-time prediction using GNNs over road network.
  4. Predictive maintenance model with scheduling optimization.
    Pricing strategy and A/B simulation for ride-hailing marketplace.

Software, libraries & frameworks

  1. Python stack: pandas, NumPy, scikit-learn, XGBoost/LightGBM.
  2. Deep learning: PyTorch or TensorFlow/Keras; PyTorch Geometric or DGL for GNNs.
  3. Reinforcement learning: Stable-Baselines3, RLlib, OpenAI Gym.
  4. Optimization: Google OR-Tools, networkx, PuLP, Gurobi/Cplex (if available).
  5. Simulation & GIS: SUMO, MATSim, OSMnx, GeoPandas, PostGIS.
  6. Visualization: Folium, Kepler.gl, Plotly.

           Datasets (public)

  1.  NYC Taxi & Limousine Commission trip records (taxi/for-hire trips)
  2.  Citi Bike / Capital Bikeshare trip data
  3. METR-LA, PEMS-BAY (traffic speed sensors)
  4. OpenStreetMap (road networks)
  5. Uber Movement (aggregated travel times)
  6. Kaggle logistics datasets (Instacart, shipping manifests)
  7. NOAA/weather, census/demographics as exogenous features
    Synthetic/simulated data via SUMO, MATSim for RL and dynamic routing

Readings & textbooks

  1. Ian Goodfellow, Yoshua Bengio, Aaron Courville — Deep Learning (selected chapters)
  2. C. Barnhart et al. — “Handbook in Transportation”
  3. “Vehicle Routing: Problems, Methods, and Applications” (edited collections)
  4. Reinforcement Learning: Sutton & Barto (2nd ed.)
  5. Selected papers / case studies (instructor to provide current papers on GNNs for traffic, RL in logistics, demand prediction, real-world industrial case studies from Amazon, DHL, Uber, Maersk).
  6. Ethics, privacy, and deployment
  7. Data privacy (location data), fairness (surge pricing impacts), safety and regulatory constraints for autonomous systems, environmental impacts include discussion and project checklist.