Testing Autonomous Robot Control Software Using Procedural Content Generation

We present a novel approach for reducing manual effort when testing autonomous robot control algorithms. We use procedural content generation, as developed for the film and video game industries, to create a diverse range of test situations. We execute these in the Player/Stage robot simulator and automatically rate them for their safety significance using an event-based scoring system. Situations exhibiting dangerous behaviour will score highly, and are thus flagged for the attention of a safety engineer. This process removes the time-consuming tasks of hand-crafting and monitoring situations while testing an autonomous robot control algorithm. We present a case study of the proposed approach – we generated 500 randomised situations, and our prototype tool simulated and rated them. We have analysed the three highest rated situations in depth, and this analysis revealed weaknesses in the smoothed nearness-diagram control algorithm.