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

91 Simulation and testing | Focus The results of these changes can be analysed using the flight control computer, with all the standard telemetry links. These approaches address perhaps 95% of the testing, and provide a higher level of reassurance that test flights will be successful. That is especially important for custom platforms, where there may be only one or two versions. Traditional simulation tools are also a key part of the development and test environment. Development tools that were previously used to develop control loops now have to include ML techniques. In the past they used a block-based graphical interface to develop the algorithms, automatically produce code optimised for a particular processor or GPU, and provide the test vectors to validate the code running on the actual hardware. This is much harder with ML algorithms, which have to be trained and with reinforcement learning where the algorithms keep taking in new data to constantly improve their accuracy. That means the algorithm development tools are moving more to simulation environments, taking in external real- world or simulated data as part of the software validation process. Summary The development of autonomous systems has driven a boom in the use of simulation technologies of all kinds. Being able to model a vehicle or even part of a city with all the different elements though is a huge computing challenge. The emerging market for common scenarios for testing, and more widely available test data with the associated test vectors, will speed up the validation of controllers and sensor systems. The prospect of golden scenarios and test environments for regulatory approval is another significant potential move that would also speed up the development and testing process. All of this requires a new generation of frameworks rather than individual tools. The ability to manage huge amounts of data effectively as part of a safety- critical design process will be vital, and that will reduce the number of suppliers to a handful. The differentiation in the simulation and test process will then sit with the development of the scenarios and the efficient analysis of all the test data being produced. Acknowledgements The author would like to thank Danny Shapiro at Nvidia, Mattieu Worm at Siemens, David Pinta and Luis Garcia at UAV Navigation, Chris Hayhurst at MathWorks and Andy Luo at TotalSim for their help with researching this article. Unmanned Systems Technology | June/July 2019 CANADA OPAL-RT Technologies +1 514 935 2323 www.opal-rt.com FRANCE AVSimulation +33 1 4694 9780 www.avsimulation.fr GERMANY PTV Group +49 721 9651 0 www.ptvgroup.com SPAIN UAV Navigation +34 91 657 2723 www.uavnavigation.com UK rFpro +44 2380 989 235 www.rfpro.com USA Cognata +1 718 317 8900 www.cognata.com dSpace +1 248 295 4700 www.dspaceinc.com MathWorks +1 508 647 7000 www.mathworks.com Metamoto - www.metamoto.com National Instruments +1 877 387 0015 www.ni.com Nvidia +1 408 486 2000 www.nvidia.com Realtime Technologies +1 248 548 4876 www.faac.com/realtime-technologies Siemens PLM +1 800 498 5351 www.plm.automation.siemens.com Synopsys +1 650 584 5000 www.synopsys.com TotalSim USA +1 614 255 7426 www.totalsim.us Some examples of simulation & testing providers for autonomous vehicles