One of these models, Holyhead Road, feeds into the centre of the city of Coventry (UK) which is also part of the MFM testbed. The selection of routes in this digital model present an excellent representation of the varied types of urban roads which can be found across cities within the UK. Due to the complexity of the potential traffic scenarios and layouts in this model, there is huge scope for generating a massive variety of edge cases (unusual, unlikely, but possible situations which need to be simulated via artificial intelligence so that the vehicle is prepared to handle any situation that occurs). This makes Holyhead Road an ideal model for synthetic data generation.
The location includes extensive signage which have been modelled to really challenge algorithms. For example, some with be obscured with dirt or are void of information (non-road signs!). Tree-lined roads also cast multiple shadows over roads and objects and can therefore cause potential issues for sensors and the related algorithms.
This location includes a wide range of features:
- Multi-lane urban roads
- A complex multi-lane traffic light-controlled junction
- Merging roads
- Traffic light controlled pedestrian crossings
- Classic urban 2-lane roads
- Cycle lanes and bus stops
- A large complex roundabouts with five exits
- A smaller circular four-exit roundabout
- Multiple classical T-junctions
- Pedestrian traffic island crossings
- Multiple parked cars
- Petrol Station
When populated with traffic, this model has the capability to deliver a busy environment which will challenge perception systems, particularly when navigating the more complex layouts.
There are two significantly complex sections within our Holyhead Road model: a five (yes, that’s FIVE) exit roundabout and a multi-lane traffic light-controlled junction. The model includes a considerable amount of visual complexities, not only road furniture, but also surrounding locations, residential areas and a variety of signage.
Autonomous vehicles can not be feasibly validated in the real world alone as there simply isn’t enough time or resource to achieve this. The industry must also exploit the expandability of simulation and this model provides a varied selection of roads, common to the UK, upon which to undertake AV testing and training.