How Mobile Lidar is Revolutionising Automotive Testing

A 1.929Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

rFpro has been a customer of the mobile lidar industry for the last seven years. Our organisation rFpro delivers driving simulation solutions to automotive and motorsport manufacturers around the globe. Over the last 7 years we have commissioned over 200 kinetic lidar surveys of race tracks for F1, NASCAR and LMP as well as test tracks, proving grounds and thousands of kilometres of public road. The work has always been interesting and, on occasion, challenging.

The end-users of our solutions range from design teams at top Formula One groups to engineers at major car manufacturers creating the latest Autonomous Driving and Advanced Driver Assistance Systems (ADAS). The common thread linking our users is that they share a common requirement to reproduce the real world, with a surprisingly high level of detail and accuracy.

Possibly of more interest to LiDAR News readers is the fact that lidar technology has been instrumental in transforming driving simulators. Until about seven years ago driving simulators were mainly used for what we call procedural simulation; driver training, evaluating human factors such as fatigue and stress, ergonomics and testing new man-machine-interfaces. What lidar brought was an accurate map of the real world and with it the possibility to simulate a mathematical model of a car, over exactly the same road surface, with identical scenery and visual reference, with a human driver, in a safe, controlled, environment. It actually took us another two years of R&D to develop a tool, TerrainServer, that is capable of feeding 1cm resolution lidar data into the simulated vehicle models fast enough that they can run real-time.

Driving simulators come in all shapes and sizes. Our business focuses on the fastest, highest bandwidth systems targeting engineering development. They are designed to accurately replicate the dynamic behaviour of the vehicle being simulated. When measurement equipment fitted to a real car that is on a lap of the Nordschleife test track in Germany, records a disturbance in the road surface due to a repair or joint in the road surface, we have to be able to replicate that same discontinuity in the simulated car. This means that the demands we make on our suppliers in the mobile lidar industry have often been rather challenging.

The Nordschleife, mentioned above, is probably the most iconic test facility in the entire automotive world. Ex-F1 driver Jackie Stewart referred to it as the "Green Hell", 20.8km of twists and turns create a very demanding test track. Our digital model of this track is a good example of the challenges we face. The entire road surface, all 20.8km, has been mapped to 1cm resolution with sub-millimetre accuracy (using phasebased lidar for the road surface) and we have to feed this in real-time to our end users’ vehicle models.

The vehicle models are mathematical representations of their entire cars, which typically calculate the motion of every component of the chassis and drivetrain 1000 times a second, allowing us to replicate the car’s dynamic behaviour, on a simulator, with a very high degree of correlation. This is challenging not just for our lidar partners to produce a point-cloud that has a density in excess of 1 point per square centimetre, with sub-millimetre precision–but also for us to manage that volume of data, in real-time, integrating every single point beneath each of the four tyres’ contact patches 1000 times per second.

Interestingly, relative accuracy is not as important to us as repeatable precision. A 1cm error between two points 100 metres apart does not really matter–across that distance we will see an error in the gradient of just 0.01%. However for two adjacent points, 1cm apart we need their relative position to be measured with precision. If noise in the point-cloud exceeds the magnitude of the disturbances in the road surface that we are aiming to replicate then the data has no value.

The most challenging applications are those targeting ride-comfort and durability studies. These exercise the upper frequencies of the systems being modelled on the car and require a very detailed road surface to be fed into each tyre contact patch. It is immediately apparent that this will be of interest to manufacturers of passenger cars. What is less obvious is that it is also important to Formula 1 and NASCAR teams; every single bump and ripple in the tarmac is feeding energy into the system, affecting the tyres’ temperatures and the behaviour of the dampers. For these ride and durability studies, in order to achieve the data quality required, we are into the realm of phase-based lidar sensors.

Not all applications are so challenging. The demands are reduced if the end-user is less interested in ride-comfort. For example they may be focusing on handling behaviour, either a sports car performing at the limit, or they are evaluating the performance of a car in emergency manoeuvres, perhaps to test new stability control electronics or active safety systems. In either case the data requirements are still well within the realm of high-quality mobile lidar systems. We will be looking for a point-interval of just 5cm with a level of precision of <2mm. In practice this means we still need to be working with partners using high-end IMUs, however the sensors can be TOF (Time Of Flight).

TOF-based systems deliver better results for us is when we’re modelling public roads to support the development of autonomous driving and ADAS systems. The greater range of TOF sensors is a significant help to us because more of the road-side scenery is captured. This helps us create a highly realistic copy of the real world with all the detail that the driver needs in order for the environment to be a realistic facsimile. When building these digital circuit models we also use the road-side furniture, parsed from the lidar data, to define road networks, for example the painted lane markings, road signs, overhead gantries and junctions. These will be used to help control simulated traffic, other cars and trucks that are added to the simulated world, to create a more realistic test environment.

The scans we have commissioned have been used to create digital models that are helping our customers develop the chassis, drivetrain and electronic control systems earlier in their car’s development lifecycle and without the expense of building prototype test mules. The cost savings are immense. A high-end vehicle dynamics simulator might cost around $2-3m and some of our customers have recovered those costs on their first project. Consequently the rate of adoption of driving simulator technology is increasing and we are predicting that the demand for high quality mobile lidar from our industry will continue to grow until the entire world road network has been scanned, to survey grade, with sufficient density to meet the needs of the automotive manufacturers.

We have closely watched the lidar technologies our partners have been using develop over the last seven years. Our belief is that the software solutions for processing large lidar datasets has lagged behind the hardware and we’re only now starting to see quality off-the-shelf solutions for software we had to create in-house. In the short term we’re looking forward to faster scanning solutions that can map the world’s road networks to the density and accuracy required, while driving at normal road speeds, even in challenging urban environments with poor position data.

Longer term we wonder if cost considerations will drive the industry towards a crowd-based approach to scanning. By aggregating data from millions of car-mounted ADAS sensors, geo-referenced to previously mapped routes, the manufacturers will ensure continuous updating of the road network. At the moment this sounds like fantasy but the approach is already working in the marine world, where Navionics navigation charts are augmented from a crowd-sourced bathymetry map using the sonar data of every vessel recording their depth data.

rFpro was founded in 2007 to bring to market a new type of driving simulator focused on engineering development. Chris Hoyle is rFpro’s founder and Technical Director.

A 1.929Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE