Logistics & Operations

The original VRP paper; introduced linear programming formulation for fleet routing from depot to customers.

Introduced the 'savings algorithm'—still the most widely used VRP heuristic in commercial routing software.

Uber research on matching algorithms and dynamic pricing; batching and bipartite matching for real-time dispatch.

Authoritative survey from Dantzig-Ramser through modern metaheuristics.

Google's open-source production solver powering Cloud optimization APIs; handles complex industrial constraints at scale.

Industry-standard LKH heuristic holding records on all large benchmark instances; extended as LKH-3 for CVRP variants.

Production RL system matching tens of millions of daily rides; won NeurIPS 2018 Best Demo Award.

Foundational exact method tutorial underpinning state-of-the-art solvers for fleet planning.

Production algorithms enabling 30-minute grocery delivery across Alibaba subsidiaries in China. Franz Edelman Award finalist.

First large-scale public real-world routing dataset (9,184 routes); catalyzed ML + TSP research combining driver tacit knowledge.

Introduced 'echelon inventory' concept; proved optimal base-stock policies for serial supply chains.

Definitive textbook on newsvendor, (s,S) policies, and dynamic inventory systems.

Guaranteed-service model for safety stock optimization; deployed in SAP, Kinaxis, and enterprise software.

Amazon's multi-echelon system with Lagrangian decomposition for real-time optimization across 175+ fulfillment centers.

Deep learning + simulation-optimization generating $42M+ annual savings in inventory and shrinkage costs. Franz Edelman Award finalist.

End-to-end algorithm improving local fulfillment rates by 0.54% across millions of weekly orders in two-echelon distribution. Franz Edelman Award finalist.

State-of-the-art picker routing for mixed-shelves warehouses; exact real-time algorithms for high-SKU e-commerce facilities.

Foundational Amazon paper powering SageMaker and Amazon Forecast; ~15% accuracy improvement over classical methods via global RNN.

Landmark Walmart competition (42,840 time series) proving LightGBM ensembles outperform traditional statistical methods at retail scale.

Companion competition on probabilistic forecasting; winning methods combined LightGBM + DeepAR for intermittent demand quantiles.

Attention-based architecture achieving 7% lower P50, 9% lower P90 losses; combines performance with interpretable variable importance.

GraphDeepAR integrating GNN encoder with probabilistic decoder; captures inter-article relationships for improved retail forecasting.

Deployed GNN for Google Maps ETA; reduced negative outcomes by 40%+ in major cities.

Uber's production ETA using Transformers for tabular data; highest QPS system at Uber.

Foundational spatiotemporal traffic forecasting combining graph convolutions with sequence models.

Deployed serving 100+ million packages/day; achieved 13.8% RMSE improvement using spatial-temporal LSTM with attention.

First joint route-and-time prediction for food delivery; transformer architecture handling 34.9 million orders/day on Meituan.

Heterogeneous graph neural networks modeling road networks as multi-relational graphs; establishes importance of graph representations.

First data-driven framework predicting delivery time distributions (not point estimates); tested on JD.com showing 6.1% sales improvement.

Comprehensive survey covering Cainiao, JD, Meituan, GrabFood systems; taxonomizes by task type, architecture, and learning paradigm.

Definitive survey on call center operations; directly applicable to gig economy capacity planning.

Foundational heavy-traffic theory; introduced QED regime balancing service quality with utilization.

Natural experiment (NYE outage) showing dynamic pricing equilibrates supply/demand in real-time.

Studies driver labor supply response using 25 million Uber trips.

Foundational theory showing surge pricing achieves near-optimal profits with self-scheduling workers; directly informs Uber/Lyft design.

Rigorous econometric study showing incentive reallocation can increase capacity by 22% without additional cost.

Characterizes optimal spatial pricing; introduces 'balancedness' concept for demand patterns and driver positioning incentives.

Developed at Uber showing surge pricing solves 'wild goose chase' matching failures better than matching adjustments alone.

Proves multiplicative surge is not incentive-compatible; proposes additive surge mechanism deployed by Uber.

Rigorous queueing-theoretic model; develops fluid-based max-weight matching achieving √η optimality rate.

Authoritative survey on DoorDash, Instacart, Uber Eats covering matching, pricing, and gig economy logistics.

Comprehensive review classifying platforms by scheduling mechanism.

Reduced drive distance by 62% and saved $500M+; demonstrates OR impact at scale.

Foundational FSTSP model for truck-drone tandem delivery; inspired by Amazon Prime Air, Google Wing, and DHL drone programs.

Extends to multiple heterogeneous drones; MILP and heuristics for 100+ customers with 4+ drones.

Rolling horizon optimization for Amazon Flex-style platforms; shows 37% vehicle-mile savings vs. dedicated fleets.

Seminal paper on tactical time-slot design for e-grocers; developed with Albert.nl; 237+ citations in last-mile literature.

First analytical study of crowdsourcing with guaranteed time windows; robust queueing for Amazon Flex-style operations.

Mixed long/short delivery windows; tested with German e-grocer AllyouneedFresh data.

THE Kiva paper by its three co-inventors; describes multi-robot coordination that Amazon acquired for $775M and scaled to 500,000+ robots.

Highly-cited queueing network model for Amazon-style RMFS; foundational for warehouse layout optimization and throughput analysis.

THE canonical survey (3,000+ citations) covering layout, storage assignment, routing, batching, and zoning.

Simulation-based evaluation of practical decision rules for order assignment and pod selection in Kiva-style warehouses.

Survey of 26 teams in Amazon's manipulation challenge; synthesizes perception, motion planning, and grasp planning lessons.

Modern survey covering order batching, wave planning, RMFS, scattered storage, and returns processing for e-commerce.

THE foundational EVRP paper (1,000+ citations) introducing charging constraints into last-mile routing; established benchmark instances.

Foundation for green vehicle routing (1,500+ citations); models CO2 as function of speed, load, distance for cost-environment tradeoffs.

Addresses practical fleet electrification decisions; optimizes routing for mixed fleets with realistic energy consumption models.

Franz Edelman Award winner; industry-deployed system saving $300-400M annually, reducing 100M miles, 10M gallons fuel, 100,000 metric tons CO2.