Analyzinf the Quuality of Sensor Data with Simulations combined with Automated Theorem Proving

Abstract

Self-Driving vehicles are still in the development process and will soon be part of our everyday life. There are companies working with this technology today and have already demonstrated a prototype of those self-driving vehicles, one of those companies is Google. Over the years ideas have been spread around in the world and many developers wanting to be part of the new technology. The DARPA Grand Challenge was created to gather skilled developers from around the world to compete with their automated cars. In this paper we focused on the efficiency part in automated parking by studying the sensors mounted on and around the vehicle. The sensors will be analyzed systematically by injecting noise data and also skipped sensor data. The vehicle will be tested with different parking scenarios in a simulating environment and the outcome of the tests will be verified by using an Automated Theorem Prover called “Vampire Theorem Prover” to draw conclusion according to the results. To determine the ground truth, we ran 100 test with different parking scenarios from which we got a subset of 58 scenarios at which the car parked successfully according to the specification while using 100% sensor quality. Selecting ten scenarios from the ground truth, we ran the tests with different noise levels and observe the parking accuracy. To achieve a parking accuracy of 90%, the sensor(s) used should have about 90% quality.