Advanced simulation aims to remove testing barrier for autonomous vehicle development

The simulation technology enables the tuning of sensors, the training of perception and control algorithms and testing of the full AV technology stack.

The company said that the platform can provide closed-loop perception testing and the creation of engineering-grade synthetic training data for the autonomous vehicle industry.

“AV elevate is a game-changing simulation platform, it is the most advanced solution to enable the complete AV technology stack to be tuned, trained and tested,” Matt Daley, technical director at rFpro, said in a statement. 

“For the first time, AV developers can confidently reduce their reliance on real-world testing, instead subjecting systems to AV elevate’s highly-accurate synthetic data. Now, testing and development can be massively and cost-effectively scaled like never before, removing the biggest barrier to the advancement of AVs.”

To create AV elevate, rFpro has integrated several new technologies to its existing platform including LiDAR, radar and camera models, a new Simulation Manager to simply define the full vehicle sensor suite and create base test scenarios with thousands of iterations, and compatibility with High Performance Computing (HPC) in the cloud to conduct and scale the testing rapidly.

Tuning sensor systems

AV elevate integrates high-fidelity sensor models for all major AV sensor types and enables installation choices and configurations to be tuned and optimised. The synchronous platform allows for hundreds of sensors to be tested, enabling sensor fusion testing to a level of accuracy not previously possible, the company said.

Included within the simulation solution is a library of standard sensor models alongside digital twins of commercially available sensors. rFpro said that this allows development to progress before a physical sensor exists or enables OEMs to benchmark technologies against their competition.

Automating annotations

Typically, AV developers manually annotate each frame of video, LiDAR point or radar return to identify objects in the scene to create training data. According to rFpro, this approach generally takes 20 minutes per frame and has a 10 per cent error rate.

Instead, AV elevate automates this process using engineering-grade synthetic training data, meaning it is ‘100 times faster’ and ‘150 times more cost-effective’ than manual annotation, and error-free, the company said.

Changing scenarios

rFpro’s Simulation Manager can automate variations with both the vehicle model and the environment. These include changes to the sensor types, positioning of sensors on the vehicle, traffic, pedestrians, time of day, weather conditions, street furniture and obstructions.

It can quickly enable the creation of focused variations of the base scenario to be generated, creating hundreds of edge case scenarios for testing. Users can create their own database of scenarios or connect to large external third-party databases.

At the core of AV elevate is rFpro’s physically modelled virtual environments and ray tracing rendering technology. Its library of over 180 real-world digital twins provides a virtual proving ground with every element in the scene physically modelled with realistic material characteristics and a road surface model accurate to 1mm.