Road-Surface Modeling Aims to Support Autonomous Driving

A road can be sampled every 5 mm across its surface with sub-one mm precision using rFpro's new process.

Some chief executives are enthusiastically fond of stating that their companies already possess much of the technology that would, in theory, allow them to introduce autonomous vehicles in the very near future. However, there is at least one area that is likely to require not just in-depth knowledge but, quite literally, superficial knowledge.

It is to acquire a total understanding and appreciation of road surfaces that would allow an autonomously operated vehicle’s intelligent systems to recognize and appreciate the potential effects on a vehicle of anything from mild camber changes and potholes on a minor road, to cracks on a freeway, and to ensure that none would cause a possibly catastrophic effect that could endanger a whole high-speed convoy.

Regular image of the road surface sampled.

To achieve this will involve very comprehensive test programs incorporating the use of advanced surface scanning technology of roads in all markets served by an OEM, said Chris Hoyle, Technical Director of simulator software specialist rFpro. The company, which has technology links with Ferrari's F1 operation (see Ferrari Changes Course on Track Simulation Software), is majoring on the creation of solutions to the challenge. And it is doing so by combining accuracy with new time saving systems.

“The needs of autonomous driving test programs are only now starting to be fully understood,” said Hoyle. “But meeting those needs can be achieved via expertise that can be gained now, providing equally significant information to make current vehicles safer.”

Hoyle explains that as increasing numbers of new vehicles are designed to suit global markets, making the optimum design choices at the outset has become essential but is now far more challenging for vehicle manufacturers. Revisions to a vehicle to suit individual market tastes and conditions late in a development program are so costly that manufacturers aim to avoid such issues by testing as early as possible.

Pronounced regional differences in road-surface characteristics are a known headache for use of global platforms because effective testing is currently only possible after representative prototype vehicles are available, stated Hoyle. “Driver-in-the-loop (DIL) simulators provide a solution by allowing a human driver to experience the vehicle’s behavior in a virtual environment long before physical prototypes exist; however, to be effective they need a level of road-surface detail that has been unavailable until now.”

However, Hoyle says his company has developed a solution: “New surface scanning technology is being utilized by rFpro to produce digital road models with unprecedented accuracy and speed. We can capture and reproduce the differences between a frost-damaged Detroit highway, a smooth German autobahn, a Swiss alpine pass, or any other surface.”

This advance means vehicle manufacturers located anywhere in the world can evaluate their vehicle’s chassis response to any road type in a realistic virtual environment, he claims: “And they can do so without leaving the office.”

The breakthrough in digital road modeling developed by rFpro starts with replacing the usual single pulse laser LIDAR “time-of-flight” scanning process with new scanning technology that uses a number of separate, phased laser signals. Instead of waiting for each signal to return before firing the next one, the controlled phasing allows the signals to be overlapped, increasing the speed, quantity, and quality of data captured.

“The new process provides up to 50 times the level of detail with greater accuracy than ever before; it’s also faster, which allows the scanners to drive at normal road speeds rather than at a crawl,” said Hoyle. “This makes it realistic to scan much longer sections of a chosen route, even during the day, without impeding other road users.”

To provide global coverage, the company works with a core group of regionally located scanning partners, all of whom have now added the latest phase-based scanning capability to the systems they use for rFpro’s road surveys.

Chris Hoyle, rFpro Technical Director, claims the new process provides up to 50 times the level of road detail with greater accuracy than ever before.

Because it reduces development time, the new technology has the potential to make a significant reduction in costs.

Accuracy is such that a road surface can be “sampled” every 5 mm (0.2 in) with a precision down to less than 1 mm (0.04 in).

Growing numbers of vehicle manufacturers are asking for their favorite test routes to be captured and reproduced, according to Hoyle: “For example, in 2012 we built just over 100 km of digital road models; in 2014, it was more than 1000 km, and this year we expect to build approximately 3000 km.”

He added that the increasing volumes of data have led rFpro to switch to cloud-based processing and storage, providing almost limitless scope for further growth and enabling hundreds of CPU cores to work simultaneously on data processing, further reducing timescales: “Traditionally, manufacturers have used simulators adapted from the aerospace industry, but to get the most from these digital road models it is essential to use state-of-the-art, purpose built automotive simulators.”

This year, rFpro has begun working with several major OEMs on advanced driver assistance systems for autonomous vehicles, although Hoyle will not comment on the likelihood of the commercialization of autonomous vehicles by 2020, a timescale for which that some companies are aiming.

Current developments in the automotive industry, such as the growing use of shared platforms and autonomous vehicles, are increasing the importance of virtual testing.

The company’s TerrainServer software is currently being used by OEMs in North America, Asia, and Europe as well as by most of the F1 and NASCAR teams, stated Hoyle. It takes the digital track (or road) surface model together with other inputs including tire contact patches, and processes the information in real time, to achieve accurate vehicle dynamic simulations. Hoyle explained that the real-time aspect of this capability is essential for DIL testing.