Performance Boundary Identification for the Evaluation of Automated\n Vehicles using Gaussian Process Classification

2019/07/11 by Felix Batsch, Batsch, Felix, Alireza Daneshkhah +7 · 1 citation
Engineering · Computer Science · #Autonomous Vehicle Technology and Safety #Gaussian Processes and Bayesian Inference #Vehicle emissions and performance

paper · pdf · doi:10.48550/arxiv.1907.05364

Abstract

Safety is an essential aspect in the facilitation of automated vehicle\ndeployment. Current testing practices are not enough, and going beyond them\nleads to infeasible testing requirements, such as needing to drive billions of\nkilometres on public roads. Automated vehicles are exposed to an indefinite\nnumber of scenarios. Handling of the most challenging scenarios should be\ntested, which leads to the question of how such corner cases can be determined.\nWe propose an approach to identify the performance boundary, where these corner\ncases are located, using Gaussian Process Classification. We also demonstrate\nthe classification on an exemplary traffic jam approach scenario, showing that\nit is feasible and would lead to more efficient testing practices.\n

Cited by

• Efficient Safety Testing of Autonomous Vehicles via Adaptive Search over Crash-Derived Scenarios

Related

• Identification of Challenging Highway-Scenarios for the Safety\n Validation of Automated Vehicles Based on Real Driving Data
• Exploring the Boundaries: Thorough Software Testing for Safety-Critical Driving Scenarios Based on Kinematics in the Context of Autonomous Driving
• Learning Driver Behaviors Using A Gaussian Process Augmented State-Space Model
• Training Adversarial yet Safe Agent to Characterize Safety Performance of Highly Automated Vehicles
• Evaluation of Automated Vehicles in the Frontal Cut-in Scenario - an Enhanced Approach using Piecewise Mixture Models