Building Understanding of Modern Last-mile Delivery Systems
5 likes1,279 views
1. Same-day delivery systems that provide instant delivery are becoming more common and changing consumer expectations and retail logistics. 2. Very short delivery times require new approaches for efficient last-mile delivery logistics. 3. Understanding modern last-mile delivery systems can be helped by quantitative models, but more study is still needed as these systems evolve rapidly.
Building Understanding of Modern Last-mile Delivery Systems
- 1. Building Understanding of Modern Last-mile Delivery Systems Alan Erera School of Industrial and Systems Engineering, Georgia Tech Transportation Symposium to Honor Carlos, Berkeley, June 2018
- 2. The impact of an advisor • Inspire with guiding principles • Before seeking solutions, be sure to understand the problem • Arm with diverse toolkit • “I can teach you things that no one else can!” • Coach but also cheerlead • Root for your students and prioritize them professionally and personally • Be a role model • The impact of an advisor is primarily delivered through her students
- 3. The impact of an advisor Team Carlos Team Erera
- 4. Last-mile delivery logistics collaborators • Georgia Tech collaborators: • Alex Stroh, Ramon Auad, Damian Reyes (Amazon), Mathias Klapp (PUC Chile) • AE, Alejandro Toriello, Martin Savelsbergh • Industry collaborators • Grubhub, SF Express
- 5. What to remember 1. Same-day delivery systems with short delivery deadlines are here to stay, providing consumers instant acquisition and large selection 2. Very short click-to-door times may require different approaches for last-mile delivery logistics to consumers 3. Building an understanding of modern last-mile delivery logistics systems can rely on a variety of different quantitative models 4. Much work remains to be done in the study of evolving delivery logistics systems
- 6. What to remember 1. Same-day delivery systems with short delivery deadlines are here to stay, providing consumers instant acquisition and large selection 2. Very short click-to-door times may require different approaches for last-mile delivery logistics to consumers 3. Building an understanding of modern last-mile delivery logistics systems can rely on a variety of different quantitative models 4. Much work remains to be done in the study of evolving delivery logistics systems
- 7. Fundamental change in commerce • Diminishing role of retail stores • Delivery of product directly to consumers • Product “definition” includes a service element • When and how can I get it? • Last mile logistics • Increasing complexity!
- 8. Same-day delivery retailing • Order (click) and receive (to door) • on the same day • within a few hours • within an hour • Best of both worlds • No trip to store • Much larger product catalog than store
- 9. What to remember 1. Same-day delivery systems with short delivery deadlines are here to stay, providing consumers instant acquisition and large selection 2. Very short click-to-door times may require different approaches for last-mile delivery logistics to consumers 3. Building an understanding of modern last-mile delivery logistics systems can rely on a variety of different quantitative models 4. Much work remains to be done in the study of evolving delivery logistics systems
- 10. Same-day delivery • Order placement and ready time processes Ready time r time today’s orders begin order deadline o4 ready for dispatch from depot o4 Placement time a Order placed by customer “click time”
- 11. Same-day delivery • Order placement and ready time processes ready time process realization timeo1 o2 o3 today’s orders begin on order deadline o4
- 12. Same-day delivery • Vehicle operating deadline • Order delivery deadlines (click- or ready-to-door) time vehicle operating deadline TOrder delivery deadline o1 o2 o3 o4 on
- 13. • Consolidation and vehicle dispatching • Dispatch one or more vehicles over time, serving orders by delivery deadlines Same-day delivery time Driver 1 o1 o2 o3 o1 o2 o3 Single depot system for now
- 14. • Consolidation and vehicle dispatching • Dispatch one or more vehicles over time, serving orders by delivery deadlines • Vehicles may be reused, must return from last dispatch by operating deadline Same-day delivery dispatching time Driver 1 Driver 1Driver 2 Single depot system for now
- 15. What to remember 1. Same-day delivery systems with short delivery deadlines are here to stay, providing consumers instant acquisition and large selection 2. Very short click-to-door times may require different approaches for last-mile delivery logistics to consumers 3. Building an understanding of modern last-mile delivery logistics systems can rely on a variety of different quantitative models 4. Much work remains to be done in the study of evolving delivery logistics systems
- 16. Single depot, single deadline setting time deadline T o1 o2 o3 o4 on last order N depot o1 service area A
- 17. Dedicated vehicle fleet: how many? time deadline Tlast order NContinuous arrival rate 1 per scaled time
- 18. Dedicated vehicle fleet: how many? time deadline Tlast order NContinuous arrival rate 1 per scaled time depot service area A Approximate travel time to serve n customers across service area
- 19. Dedicated vehicle fleet: how many? time TNContinuous arrival rate 1 per scaled time depot service area A Approximate travel time to serve n customers across service area One vehicle can serve all requests for if (T – N) large enough!
- 20. Large fleet: simple “whole area” strategy time TNTo minimize total duration of all routes: “wait as long as possible” depot service area A To find first dispatch time: each vehicle serves customers in “whole area”
- 21. Large fleet: simple “whole area” strategy time TNTo minimize total duration of all routes: “wait as long as possible” depot service area A To find first dispatch time: each vehicle serves customers in “whole area” To find the j-th dispatch time:
- 22. Large fleet: simple “whole area” strategy time TNTo minimize total duration of all routes: “wait as long as possible” depot service area A For some large enough m: each vehicle serves customers in “whole area” Under this strategy, you need only m vehicles: and the m-th vehicle may complete its work before T
- 23. Large fleet: simple “divide area” strategy time TN depot service area A Each vehicle accumulates N/m customers, and operates in a smaller zone of size A/m each vehicle serves zone with area A/m m vehicles Dispatch all m vehicles at time N
- 24. Large fleet: simple “divide area” strategy time TNDispatch all m vehicles at time N depot service area A Each vehicle accumulates N/m customers, and operates in a smaller zone of size A/m each vehicle serves zone with area A/m m vehicles
- 25. Single vehicle: serving more customers! time TN depot service area A We can move N closer to T, but only by reusing vehicle for multiple trips
- 26. Single vehicle: serving more customers! time TN depot service area A Dispatching Policy Minimizing Total Route Duration: Suppose that there is some enforced minimum dispatch size such that, for larger dispatches, q >= f(q). Then an optimal dispatch policy for a single vehicle is to initially wait, and then dispatch consecutively shorter duration trips without any additional waiting. We can move N closer to T, but only by reusing vehicle for multiple trips
- 27. Single vehicle: serving more customers! time TN depot service area A We can move N closer to T, but only by reusing vehicle for multiple trips Vehicle arrives back at depot for final dispatch after last order: Vehicle can complete final dispatch before deadline:
- 28. Single vehicle: serving more customers! time TN depot service area A Dispatching Policy Minimizing Total Route Duration: Using fewest dispatches D in this scheme leads to minimizing costs A simple iterative root-finding algorithm can be used to find D and its associated initial waiting time We can move N closer to T, but only by reusing vehicle for multiple trips
- 29. Single vehicle: serving more customers! time TN depot service area A Dispatching Policy Minimizing Total Route Duration: Using minimal number of dispatches D in this scheme leads to minimizing costs A simple iterative root-finding algorithm can be used to find minimum D and its associated initial waiting time We can move N closer to T, but only by reusing vehicle for multiple trips
- 30. Building understanding with other models • Simplified half-line geometry provides maximum consolidation benefit • Eliminates need for combinatorial route sequencing (TSP) timeo1 o2 o3 ono4 Distance depot
- 31. Building understanding with other models • Simplified half-line customer geography • Polynomial algorithms for single-vehicle multiple-trip dispatch optimization with known orders (Reyes, E, Savelsbergh, 2018) • Can orders be served by deadlines? • What dispatches minimize total duration of all dispatches? • Polynomial algorithms for single-vehicle multiple-trip order accept/reject with known orders • How many orders can be served? (Reyes, 2018 thesis) • Balancing orders served with route travel costs? (Klapp, E, Toriello, 2016) Distance depot
- 32. Building understanding with other models • Simplified half-line customer geography • A priori rollout policies for single-vehicle multiple-trip dispatching with dynamic and uncertain orders • Balancing orders served with route travel costs? (Klapp, E, Toriello, 2016) • Time-expanded network integer programs for multiple vehicle accept/reject policy analysis with known orders • Can simple reject schemes based on distance-from-depot perform close to complex optimal schemes? (Auad, E, Savelsbergh, TBP) Distance depot
- 33. Building understanding with other models • Full detailed dynamic programming for problems with “network” geography • Single-vehicle multiple-trip prize-collecting TSP for deterministic and a priori expected cost minimization • A priori rollout policies for single-vehicle multiple trip dispatching with dynamic and uncertain orders • Balancing orders served with route travel costs? (Klapp, E, Toriello, to appear) depot
- 34. Orders served versus routing costs time TN time TN
- 35. Orders served versus routing costs Balance Cost with Orders Served Maximize Orders Served
- 36. What to remember 1. Same-day delivery systems with short delivery deadlines are here to stay, adding instant acquisition to large selection for consumers 2. Very short click-to-door times require different approaches for last- mile delivery logistics to consumers 3. Building an understanding of modern last-mile delivery logistics systems can rely on a variety of different quantitative models 4. Much work remains to be done in the study of evolving delivery logistics systems
- 37. Questions still needing better answers • “Traditional” capacity management • What mix of dedicated versus on-demand delivery resources? • How and when to add on-demand capacity? • How to decide when adding on-demand capacity is preferable to dynamically suppressing demand? • How and when to integrate “Uber-like” on-demand delivery couriers? • How to provide “soft control” incentives to independent couriers to get capacity when and where you need it?
- 38. Massive scale, shortest click-to-door • Online food ordering, shared delivery network across restaurants • Not one or two vehicles • Not one or two goods pickup points (depots) • Variation and uncertainty in operating conditions • order processing (click-to-ready) • travel and service times • order demand rates • capacity availability
- 39. Questions still needing answers • Large-scale urban pickup-and-delivery systems • How to scale high velocity last-mile systems to large numbers of orders? • How to enable simple coordination of multi-layer last-mile systems? • How and when to integrate new automated technologies (both storage and transport) into large-scale systems? When we pursue these problems, let’s not forget: Before seeking answers, understand the problem
- 40. What to remember 1. Same-day delivery systems with short delivery deadlines are here to stay, adding instant acquisition to large selection for consumers 2. Very short click-to-door times require different approaches for last- mile delivery logistics to consumers 3. Building an understanding of modern last-mile delivery logistics systems can rely on a variety of different quantitative models 4. Much work remains to be done in the study of evolving delivery logistics systems