Real-Time Vehicle Dispatching

Dynamic VRP · Rolling Horizon Heuristic

A same-day delivery service receives 50–200 new orders per hour. Each must be assigned to an available vehicle within seconds. Myopic assignment wastes 15–25% in excess distance vs. intelligent batching and look-ahead dispatch.

Operational Context

How real-time order arrivals drive dynamic dispatch decisions

Dynamic Vehicle Routing

In a same-day delivery operation, customer orders arrive continuously throughout the day. Unlike static VRP where all orders are known in advance, the dispatcher must make assignment decisions under uncertainty about future demand. Dispatching a vehicle immediately serves the current request fast but may miss opportunities to consolidate nearby orders that arrive minutes later.

The fundamental tension: responsiveness vs. efficiency. Assigning instantly minimises individual response times but increases total distance. Batching requests into windows allows better route construction but delays service. The rolling-horizon heuristic re-optimises at each event (new order or vehicle becoming free), balancing these trade-offs with a look-ahead window over anticipated near-future demand.

The Problem

Modelling dynamic dispatching as event-driven re-optimisation

Decision Chain

Network Design

Fleet Planning

Routing

Crew Scheduling

Dynamic Ops

Model Component	Dispatching Interpretation
Decision epoch	Each new order arrival or vehicle becoming idle
State	Vehicle positions, current routes, pending requests, clock time
Action space	{assign to vehicle k, insert at position j, defer to next batch}
Transition	Vehicles move along routes; new requests arrive stochastically
Objective	Minimise total distance + weighted late-delivery penalties
Constraint	Vehicle capacity, time windows, max route duration

Rolling-Horizon Dispatch Formulation Minimise Σ_k distance(route_k) + λ Σ_i max(0, delivery_i − deadline_i)

// At each event epoch t:
  Given: V = set of vehicles with current positions & partial routes
           R_pending = unassigned requests
           R_forecast = anticipated requests in look-ahead window

  subject to
    load(k) ≤ Q   // vehicle capacity
    arrival(i) ≤ deadline(i) + slack   // soft time windows
    duration(route_k) ≤ T_max   // max shift length

The rolling-horizon approach re-solves a small optimisation at each event: given current vehicle states and pending requests, find the best insertion or assignment. Unlike offline VRP, the solver must run in sub-second time since decisions block dispatch. Policies differ in how aggressively they batch versus how quickly they respond.

Interactive Solver

Simulate 60 minutes of same-day delivery dispatching

Dynamic Dispatch Simulator

Select Demand Scenario

Select Dispatch Policy

Dispatch Map (60 Simulated Minutes)

Avg Response (min)

Total Distance (km)

Requests Served

Idle %

Simulation Control

Dispatch Log

Business Impact

Cost per Late Delivery ($)

Orders per Day

Select a scenario and policy, then run the simulation.

Key Evidence

Research and practice underpinning dynamic vehicle dispatching

Interested in dynamic dispatch optimisation?

From same-day delivery to ride-hailing and field service, rolling-horizon heuristics and event-driven re-optimisation drive real-time operational efficiency. Let's discuss how dynamic VRP methods can improve your fleet operations.

Connect on LinkedIn

Disclaimer

Data shown is illustrative. Vehicle movements, request patterns, and dispatch outcomes are simplified for educational purposes and do not represent any specific delivery operator, fleet management system, or geographic service area.