Modelling carrier collaboration

A fundamental ambition of the FTC2050 project was to investigate the potential impact of carriers working together to reduce their combined carbon footprint.

Route optimisation

Many vehicle route optimisation algorithms are available but they typically only consider driving and are, therefore, not appropriate for urban parcel deliveries where the driver spends more time walking than driving (our carrier surveys indicated that the van is typically stationary for 65% of the day). To address this gap, the FTC2050 project team developed an algorithm that determines the shortest or quickest combined driving and walking route for an individual driver to serve a given set of delivery addresses. Deliveries on foot were assumed to be made using a bag of limited weight and volume capacity (e.g. 25kg and 200L).  The aim of the algorithm is to minimise the total cost of travel, measured either in distance or time, by finding the optimal:

(i) grouping of delivery addresses into clusters (where single address clusters are allowed);
(ii) driving route across the clusters where one address within each cluster is designated as a parking location; and
(iii) walking routes, one for each cluster that starts and ends at the parking location. 

The algorithm ensures that the amount of deliveries made in each cluster does not exceed the carrying capacity. 

Example driving and walking tour from/to depot (square) around delivery points

A sample input is given in the left-hand figure below which shows the location of the depot as a square and the delivery addresses as circles, and a sample output (right-hand figure) for this instance.  The output consists of the driving route shown by the solid line and the walking routes shown by the dashed lines for each of the three clusters.  The parking locations are where the solid and dashed lines meet, with the driver performing a walking tour before returning to the vehicle to drive to the next parking location. Time (or distance) origin-destination matrices for both driving and walking are required as input to the algorithm.

Portering algorithm

A heuristic ‘portering algorithm’ was developed for the FTC2050 project which models drivers dropping off parcels at selected drop-off points for subsequent pickup and delivery by porters (or by cycle couriers). The user specifies the maximum individual item and total load weights and volumes that porters may carry, with drivers delivering the remaining heavy or large items. The objective is to minimise the combined cost of drivers and porters. This gives rise to a complex routing problem that entails determining:

(i) choice of drop-off point for each parcel from among a given set of options;
(ii) optimal vehicle routes, starting and ending at depot and visiting drop-off points and addresses to be served;
(iii) optimal walking routes for porters, starting at any drop-off point, picking up a parcel load, visiting delivery addresses, picking up further loads throughout the specified working hours, ending at any drop-off point.

A penalty cost function applying to the number of porters was implemented to allow their number to be minimised.  The algorithm was designed to allow modelling of micro-consolidation at the drop-off points, in which parcels from different carriers are consolidated into combined loads.

Algorithm inputs:

(i)   Available drop-off location(s) (all locations in latitude, longitude format);
(ii)  Carrier depot location(s);
(iii) Vehicles to be used (numbers and weight/volume capacities) and, for each vehicle, a list of the individual consignment weights, volumes and delivery (or collection) locations
(iv) Portering weight and volume capacities, both for individual items and total loads, and maximum daily working hours
(v) Origin-destination driving and walking (or cycling) matrices that specify times (or any other defined generalised cost) between all pairs of locations. [Note: we used OpenStreetMap and Open Source Routing Machine data to obtain these matrices. We also manually added delivery times into the matrices: 5 minutes per consignment, for drivers and 3 minutes per consignment for porters (extra time for drivers in unloading from van and walking to and from van).]

Example solution showing:

  • 2 vans (red and cyan lines)
  • 2 porters (dotted and dashed lines)
  • 6 micro-consolidation points (dark blue diamonds – one not used)
  • consignee locations associated with two carriers (red and cyan circles)

Dashboard

Fleet telemetry data is an important element in measuring and understanding the impacts of freight traffic, however it is often available in hard-to-read formats, and its analysis and visualisation often requires specialized skills and/or software. To address this issue, the FTC2050 project developed a dashboard tool that allows users to explore last-mile logistics datasets. The tool was designed to fulfil two main aims: first, to provide a set of visualisations regarding core attributes relevant to freight traffic; and second to provide accessible, intuitive tools for the user to filter the available datasets as needed.

Screenshot of the dashboard main view:

The dashboard tool was developed using software published under open-source or public licences and is accessed through any modern web browser, therefore providing a familiar and accessible environment. It runs on a node.js application and uses the Crossfilter and dc.js javascript libraries behind the scenes to combine and filter data, while leveraging the visualisation capabilities of modern browsers (through the use of the D3.js and Leaflet javascript libraries) to chart and map the dataset as needed. Finally, the user is able to apply multiple filters on the visualisations, providing and intuitive way to interact with the dataset.

Gif animation of filtering in the dashboard:

For more information see the FTC2050 project report: developing a dashboard of last-mile freight traffic and related poster.

Data visualisations

The FTC2050 team undertook a number of parcel carrier surveys that included GPS tracking of vehicles and drivers. The following video illustrates the level of activity associated with 25 vehicle rounds undertaken in central London over three working days, based at three different depots at Southwark, Holborn and Park Royal.

Agent-based modelling

The parcel delivery process as a whole constitutes a complex system, in which multiple actors (including carriers, depots, drivers, local authorities, among others) operate within the existing framework with the aim of delivering goods to consignees. However given the complexity of last-mile freight systems, it is often hard to understand the full extent and impact any policy change would have, and furthermore, due to the system’s large scale and embeddedness within urban processes, it is often impossible to test proposed changes in situ.

In response to these limitations, the FTC2050 project developed a simulation environment using the agent-based modelling (ABM) paradigm, which aims to capture the actions and interactions of relevant actors in the parcel delivery process through simple algorithmic rules, therefore enabling experiments to be carried out in silico. As an initial step, driver behaviours are encoded into algorithmic flowcharts, which are then used to dictate the behaviour of synthetic drivers, essentially allowing them to operate autonomously within a virtual representation of London, taking into account limitations such as timed deliveries, vehicle capacity, parking, etc. Using this setup, business-as-usual conditions are captured first to verify the model; once model robustness has been established, this simulation environment provides the opportunity to test many different ‘What-if’ scenarios of proposed changes and measure the impact of each, therefore allowing for a more meaningful comparison between different options.

Watch a video of ABM in action