Geospatial data improves driverless car testing

Geospatial data specialists from Connected Places Catapult have collated a huge database featuring the characteristics of every road in the UK.

The aim is to help driverless car developers design and analyse virtual testing programmes. Each test will feature scenarios that show autonomous vehicles negotiating real-life intersections; to see whether driverless cars can operate safely around other road users.

The platform was used this spring by the CERTUS project, led by HORIBA MIRA and involving Polestar, IPG Automotive, Connected Places Catapult and Coventry University. An area east of Birmingham was chosen for the virtual testing that includes Coventry, Warwick, Nuneaton and every suburban and rural road nearby.

Project CERTUS aims to reduce the cost of testing assisted and automated driving features. Use of Connected Places Catapult’s geospatial data base promises to help improve the confidence with virtual testing, as real-life environments will be more accurately represented.

A new video produced by the project team explains that one aim of the work is to focus efforts on what to test, how to test and – crucially – when to stop testing.

The geospatial database uses Ordnance Survey road data and incorporates road traffic accident statistics across the road network , and speed limits on approaching roads.

Taking forward the driverless ambition

Self-driving cars have been trialled in the UK across a range of environments for many years, and the Automated Vehicles Act 2024 set the legal groundwork for scaling up operations and services.

Millions of miles of real-world testing is one way to demonstrate safety, but this would be both prohibitively expensive and time-consuming.

A complementary method used to demonstrate driverless vehicle safety – alongside their use on test tracks and real roads – involves virtual testing, where a driving system is interrogated using different scenarios.

But there are still significant costs associated with large-scale virtual programmes. What the Catapult aimed to do through the CERTUS project was to investigate whether geospatial data can improve the efficiency of virtual scenario-based, as well as physical, testing.

A test programme was run where 5,000 variations of a scenario were played out in simulation of a driveress vehicle navigating a junction next to a cyclist.

By using geospatial data to determine the road width and junction type at each location, test engineers were able to run scenarios that were more representative of the type of situation an autonomous vehicle in that local area would encounter.

Tailoring testing to the real world

Connected Places Catapult Senior Consultant, Tom Levermore said: “The roads in the UK are not all uniform, so it makes sense for testing to be tailored using geospatial insights to the area where autonomous vehicles might be deployed.”

In time, it is hoped that geospatial data from any area of the UK could be used to allow driverless vehicle developers to understand the risk profile for any road intersection.

“This is all about understanding what's actually on the ground,” Tom added. “Up until now, that understanding has been lacking from virtual testing.”

Database could reduce testing by a fifth

For the CERTUS project, it was estimated that a similar level of testing confidence of autonomous vehicles could be achieved with up to 20% fewer tests by using geospatial insights. The Catapult recommends that in future, multiple different test programmes should be executed and compared with and without geospatial insights to demonstrate results in a robust way. Tom adds that while this work has highlighted the potential of geospatial data in the testing of autonomous systems, there is further work to confirm its efficacy.

“We look forward to collaborating with partners looking to harness the potential of geospatial data in autonomous vehicle development and operation.”

Find out more about the CERTUS project.
Read about the range of transport work carried out by Connected Places Catapult.