Simulation of Mobile Robots in Human Crowds Based on Automatically Generated Maps

Ahstract- Mobile robots are more and more used in human environments, which means they have to navigate near the walking path’ of humans. Navigation in crowds is difficult for autonomous mobile robots because humans are unpredictable mobile obstacles that make localization difficult by obscuring sensor fields of view. Especially when there are many dynamic obstacles around the robot, localization is disturbed and navigation may fail. Another challenge is that the robot has to pay special attention to humans for safety reasons. In order for mobile robots to be used safely and reliably in the vicinity of humans in the future, new algorithms need to be developed and extensively tested. In practice, these tests are very time-consuming and expensive, especially if they are done in many different environments with a large number of humans. To reduce this workload and enable extensive testing in many different environments, we present a new cosimulation in this paper. It allows to simulate crowds in the vicinity of navigating mobile robots. For this, 3D apartments are automatically generated from over 80k residential drawings, in which robots and humans can navigate. Thus, this simulation allows to perform tests in many generated environments and thus to make statements that are less dependent on the environment. In simulated experiments with up to 15 humans in an apartment, the influence of the number of humans on the localization error as well as on the navigation is investigated and the simulation results are evaluated.