Comparison and visualization of real-time recordnings against simulated environment in development of self-driving miniature vehicles

Abstract

Testing of solutions for embedded systems has repeatedly proven to be an issue. What is a perfectly working algorithm on a testing machine may not necessarily be an optimal fix for a miniature board. When working with miniature vehicle project at the University of Gothenburg, we have experienced these issues first hand. The goal of this work is to evaluate whether we can improve implementing of embedded solutions by enhancing the testing process. The enhancement would be created from enabling to test an embedded solution directly on a real world recording, without engaging the actual hardware. In order to test our hypothesis we conducted an experiment where we devised a number of scenarios, representing different types of lane following, and performed a recording of lane following with the help of a miniature vehicle. These recordings were done within both a simulated environment and a real world physical track as well as compared. We were able to determine the differences between the recordings and whether it is negligible enough, so that the simulated recording can be deemed sufficient for testing. We have concluded that our solution cannot enhance the testing unless improved further.