Introducing the highest accuracy digital twin of Tokyo’s famous Shuto Expressway (Inner Circular route C1) specifically for vehicle engineering developers. Simulating some of the world’s most challenging roads in this way significantly accelerates the training of artificial intelligence (AI) by reducing cost and risks of collecting real world data. The virtual environment has already been adopted by major vehicle manufacturers predominately for the development of autonomous vehicles.
The 35km section of road, also known as the C1 route Expressway, is one of the most challenging stretches of city roads in the world for an autonomous vehicle to navigate. With its constantly changing road curvature and elevation, complex and densely situated junctions and a huge array of road signs and markings, it is the ultimate test of autonomous vehicle technologies. And and is the perfect way to exercise and develop such capabilities safely.
Our digital twin has been modelled using survey-grade LiDAR scan data to create a vehicle dynamics grade road surface, which is accurate to within 1mm. This is key to accurately simulate the effects of every bump, drain cover and expansion joint around the full route. The environment is not only geometrically precise but functionally accurate too with each of the thousands of road signs, markings, and roadside objects being individually classified. This is critical for the development of many ADAS and autonomous systems that rely on panoptic segmentation for their training data sets.
Collecting the volume and variety of training data needed for this type of road network would be very expensive, time consuming and potentially dangerous in the real world. Our model of Tokyo’s C1 brings this highly complex road network to development engineers and researchers, wherever they are in the world.
It enables users to add in intelligent and scripted traffic to create an almost infinite number of test scenarios in this model. The types of vehicles, their speeds, colour and density of traffic can be varied and more. The rFpro system also allows a large number of humans to drive in the model at the same time, enabling the most complex edge case scenarios to be created and recorded. This offers a cost and time-effective way of creating large quantities of usable training data to improve a vehicle’s artificial intelligence.
All our digital models are extremely versatile, enabling users to maximise their investment in simulation. Our customers can even replicate complex traffic scenarios to test how well their automatic gearbox and engine mapping perform when crawling through traffic. Importantly, this can then be correlated in the real world on the exact same piece of road.