Issue 39 Unmanned Systems Technology August/September 2021 Maritime Robotics Mariner l Simulation tools focus l MRS MR-10 and MR-20 l UAVs insight l HFE International GenPod l Exotec Skypod l Autopilots focus l Aquaai Mazu

39 Simulation tools | Focus constrained environments. Proving safe operation with Level 4 or 5 autonomy in urban environments requires an even more complex modelling environment. All kinds of different buildings, street furniture, even pedestrians, provide constantly changing reflection surfaces for sensors. This will also require a full digital twin implementation of the vehicle to get the level of accuracy required to demonstrate safe operation. Performance One simulation developer has built a HIL bench to test out the performance of sensors. It allows the sensor models to be set up with multiple modes on separate computers. A low-latency connection to the high- fidelity models allows the developer to test functions in the HIL that couldn’t be tested in the model, and a PC with multiple graphics processor units was added to the HIL system. The GPUs run high-fidelity camera models, operating in the virtual environment, to provide data for the real-world ECU. The system can be scaled up to support multiple ECUs rather than multiple vehicles. Depending on the fidelity of simulation, one vehicle with a small sensor would run on one GPU. A full robotaxi with high- fidelity models would need six GPUs, fitting into a single PC server. Each server can act as an ego vehicle, and when these are connected with the simulation environment on another server with another low-latency comms network, multiple independent vehicles can be simulated in the environment. This gets significantly closer to modelling real-world environments for Level 4 and 5 operation. Predictive maintenance The digital twin model in the simulation also enables predictive maintenance. Real- world data can be continually fed into the model to emulate the ongoing use of the vehicle. This can give more insights into the performance of the system, especially the battery pack, and allows engineers to dive down into the detail of the model to a degree that simply isn’t possible with the real system. With a higher level of abstraction, the simulation can run faster than real time, and use synthetic data extrapolated from the real-world data to model accelerated ageing of the design. This can highlight a problem with a design before it becomes apparent and allow corrective action, either during scheduled maintenance or via a software upgrade, depending on the emerging fault. This approach has been applied to swarms of UAVs to monitor their performance (see sidebar on page 36). Standardising simulation frameworks With models from sensor makers being used in different simulation tools, standards become much more important. Each simulation tool has its own application programming interface (API), and while each one will work with partners to link the tool API to the sensor model in detail, a standard API risks being generic and not powerful enough for detailed modelling. To that end, the Association for Standardisation of Automation and Measuring (ASAM), in Germany, is working on standardising simulation components for autonomous vehicle driving systems. However, in order to achieve widespread use of driving simulators for function developers, the connection between the function development framework and the simulation environment has to rely on generic interfaces. To enable easy and straightforward compatibility between Unmanned Systems Technology | August/September 2021 Simulating scenes at night using multiple sensors (Courtesy of Ansys)