Digital road models accelerate autonomous vehicle development

The attraction of simulation is that vehicle manufacturers located anywhere in the world can evaluate their vehicle’s behaviour over any road type in a realistic virtual environment, without leaving the office. This allows precise repetition of test procedures without variability in weather, traffic or road conditions, and overcomes the logistics costs of travel and the confidentiality concerns surrounding the use of prototypes. In the case of autonomous vehicles, legislation to prevent the vehicles being driven on public roads can hinder development, virtual testing avoids this constraint.

“The virtual testing of autonomous and ADAS (Advanced Driver Assistance Systems) systems is now overtaking the use of simulation for the testing of passive systems,” explained rFpro’s technical director, Chris Hoyle. “This is fuelling a huge growth in demand for digital road models of the highest accuracy. During this year we expect to build approximately 3,000 kilometres of digital road models (three times more than in 2014) as growing numbers of vehicle makers are asking us to reproduce their favourite test route surfaces.”

Many of the issues faced by ADAS and autonomous vehicle developers can be reduced by validating the control system interactions comprehensively with a driver-in-the-loop prior to installation on the vehicle. The rFpro presentation will review the challenges faced when creating digital content for virtual simulation of these systems, including the priorities for both the driver and the system developers.

Factors affecting the driver include the level of detail, latency of response, resolution, refresh rate and stereo separation, in particular their consistency and the avoidance of motion sickness. The engineering requirements drive the necessary frequency content, scanning technology and the accuracy of the topography and scenery.

rFpro’s road surveys use a breakthrough in scanning technology to capture road surfaces with better accuracy than ever before, and up to 50 times the previous level of detail. The key development has been to replace the single pulse laser LIDAR time-of-flight scanning process with 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 both the speed and quantity of data captured.

“The technique isn’t just more detailed; it’s also faster, which allows the scanners to drive at normal road speeds without impeding other road users,” said Hoyle. “We have also reduced timescales by switching to Cloud-based processing which enables hundreds of CPU cores to work simultaneously on data processing each project, and provides almost limitless scope for future growth.”

Hoyle presents his paper, entitled “Digital road models for the virtual testing of autonomous systems” at 10:15 in room B on day 2 of the symposium.