2020/02/16 by Dhasarathy Parthasarathy, Karl Bäckström, Parthasarathy, Dhasarathy +5 · 1 citation
Computer Science · Engineering · #Autonomous Vehicle Technology and Safety #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Modeling and Simulation Systems #Real-time simulation and control systems
paper · pdf · doi:10.48550/arxiv.2002.06611
openalex publication_date 2020/02/16 · openalex created_date 2022/07/19 · openalex updated_date 2026/07/28
Testing automotive mechatronic systems partly uses the software-in-the-loop\napproach, where systematically covering inputs of the system-under-test remains\na major challenge. In current practice, there are two major techniques of input\nstimulation. One approach is to craft input sequences which eases control and\nfeedback of the test process but falls short of exposing the system to\nrealistic scenarios. The other is to replay sequences recorded from field\noperations which accounts for reality but requires collecting a well-labeled\ndataset of sufficient capacity for widespread use, which is expensive. This\nwork applies the well-known unsupervised learning framework of Generative\nAdversarial Networks (GAN) to learn an unlabeled dataset of recorded in-vehicle\nsignals and uses it for generation of synthetic input stimuli. Additionally, a\nmetric-based linear interpolation algorithm is demonstrated, which guarantees\nthat generated stimuli follow a customizable similarity relationship with\nspecified references. This combination of techniques enables controlled\ngeneration of a rich range of meaningful and realistic input patterns,\nimproving virtual test coverage and reducing the need for expensive field\ntests.\n