Unmanned Systems Technology 026 I Tecdron TC800-FF I Propellers I USVs I AUVSI 2019 part 1 I Robby Moto UAVE I Singular Aircraft FlyOx I Teledyne SeaRaptor I Simulation & Testing I Ocean Business 2019 report

85 That means the verification tools are also key – the test vectors that are used are being constantly updated to ensure they provide the maximum possible test coverage. These are used throughout the design process to provide a way to distinguish between the suitability of different designs. The actual chips, boards and embedded software can then be used instead of the design models, and added as hardware in the loop in the data centre. Again, the test vectors ensure that the final integration of the hardware and software accurately reflects the performance predicted by the original behavioural models. The power of this approach is that it can be extended to wider systems: models of vehicles can be implemented in a digital model of a city. The vehicles can travel though the environment, interacting with each other and models of other road users and pedestrians, with software instrumentation allowing visibility of data that would otherwise be hidden in the system. These digital models can be connected by models of wireless links that interact with models of the buildings, trees and street furniture to test out the connectivity, allowing the effectiveness of protocols such as IEEE 802.11p, 4G and 5G wireless to be assessed. All of this is possible without having to build a single board. The advent of low-cost cloud computing allows all these models to be combined. One challenge though is that the models exist across many parts of the supply chain. Some chip makers will have the behavioural models and high-level designs of the graphics processing units (GPUs), which act as domain controllers around the vehicle as well as running the machine learning frameworks for image identification and sensor fusion. Others will have the models of processors or wireless transceivers. Tier One suppliers will have models for the sensors, platform electronics, printed circuit boards and embedded software. The OEM car maker will have the detailed model of the vehicle, from the powertrain to the modelling of the passengers and the environment they are travelling in. Other suppliers are now delivering city-level models that allow for traffic flow optimisation, analysis of noise and many other factors. These can use software engines to model roads, trees, other cars and pedestrians to create traffic scenarios. They are all physics-based, using the reflective properties of the models to see how the radar, camera, Lidar, ultrasound and other sensor models respond. For example, the model of a radar sensor will include a model of the emitter, a model of how the radio waves are reflected by wet or dry trees, and what happens with multiple reflections – all within the digital twin. One way to manage the complexity of such digital twins is not to run them in real time. As the interactions are all synchronised, a scenario can be played out and replayed, tweaking different parameters in each model. This can use different levels of validity to explore particular areas of interest. To test sensor fusion in a controller, for example, a simpler parameterised model of the radar can be used. To test the performance of the radar, other models can simulate the electromagnetic impact of the paint on the bumper of a vehicle on the radar signal. Different paint colours have different effects, so as a result Simulation and testing | Focus Unmanned Systems Technology | June/July 2019 A digital twin of a driverless car has a number of layers of models, from the chip through the vehicle design to the comms network (Courtesy of Siemens) Detailed models have been built of sections of a highway in California to create scenarios for simulation environments (Courtesy of Nvidia)

RkJQdWJsaXNoZXIy MjI2Mzk4