Research

Some parts of a routing system you can judge by using them. If a plan saves you a vehicle, you know. Other parts are harder to check from the outside — when Optrak says an electric vehicle can finish a round in February with a full load, you are being asked to trust a number that won’t be proved wrong until the vehicle is stranded.

Where that is the case, we would rather show our working. This page is where the published research behind those numbers lives: the papers, who wrote them, and what each one actually does in the product.

Where We Work With Universities

The University of Cambridge — Centre for Sustainable Road Freight. Optrak’s estimate of how much energy a vehicle uses on a leg comes from a method developed at Cambridge, published in Energies in 2021. It works out consumption from the physics of the journey — the weight being carried, the gradient, the speed profile, the weather — rather than from a single range figure off a spec sheet. That matters because real range varies by roughly a factor of two across those conditions. It is the basis of what the optimiser uses when it decides whether an electric vehicle can complete a round.

Lancaster University — STOR-i Centre. Work with STOR-i produced a method for building rounds with very large numbers of stops — commercial waste collection, where a single urban round can run to several thousand bins, through congested streets with time-restricted access. It is in the product as one of the ways Optrak builds an initial plan, used where the work falls into clear geographic clusters. It was presented at UKCI 2024 and published by Springer.

Nottingham University Computer Science + Lancaster University A 3-year knowledge transfer project laid the foundation for our later developments of parallel optimisation techniques, auto-tuning and hyper-heuristics.

Other Universities

In most years Optrak sponsors MSc research projects (typically 3 months). These allow us to investigate specific corners of the routing problem. Examples include auto-tuning, strategic EV fleet selection (battery size vs vehicle weight capacity) (Edinburgh University) and the routing impact of length of time-windows (Heriot-Watt).

Work We Build On

Vehicle routing is an active research field. Hundreds of papers are published every year, and the methods in them are public — free for anyone to implement. Optrak reads that literature and takes what works. Two recent examples now in the optimiser: Slack Induction by String Removal (SISR), published in Transportation Science in 2020, which removes whole strings of nearby stops from a plan so they can be put back somewhere better; and crossover, an idea borrowed from genetic algorithms, which takes two good plans and builds a new one from the strongest parts of each. The classics are in there too — Clarke-Wright savings, from 1964, is still a fast way to build a first plan.

The mix is ours. These methods were published as stand-alone algorithms, each tested on its own. Optrak runs them side by side — typically 10–14 at once on a desktop machine — over a shared pool of candidate plans, so an improvement found by one method is immediately available to every other. Parallel heuristics describes how that works, and the heuristic catalogue lists the full set of methods.

The second difference is the problem itself. Tim Pigden’s 2013 paper below argues that the standard academic description of the routing problem leaves out the things real fleet operations depend on — orders as opposed to a simple quantity, drivers and tractors and trailers as separate things to schedule, and loading that a single capacity number cannot describe. The methods are well-known; what they are pointed at is not.