Source Themes

This paper addresses challenges in testing Automated Driving System (ADS) functions, proposing a novel medium-fidelity evaluation and modeling approach. It aims to scale testing processes and improve test case interpretability, given the complexity of ADS setups. The proposed framework uses a statistical model to handle uncertainty and errors in perception systems, encoding error propagation in space and time. Operating conditions, including hardware, software configurations, and safety parameters, are described and evaluated with real-world and simulated data. The paper offers a promising solution for effective ADS testing and continual learning.

This paper presents a novel framework for autonomous vehicles’ left-turning at intersections. It accounts for complex interactions with human drivers and pedestrians by formulating the problem as a mathematical game. Realistic modeling of diverse traffic behaviors, extracted from naturalistic driving data, enables accurate estimation and interaction, leading to effective and human-like decision-making for autonomous vehicles in intersection scenarios.

This study addresses statistical inference for change point detection using the PRUTF algorithm. It introduces methods for computing p-values, constructing confidence intervals, and proposes strategies to improve their precision. Evaluation is performed on real and simulated data.

This paper introduces PRUTF, a method using trend filtering for change point detection in piecewise polynomial signals. PRUTF offers a dual solution path for efficient stopping rules and consistent pattern recovery, even in the presence of consecutive change points. Its effectiveness is demonstrated across various signals and compared with state-of-the-art methods using real-world datasets.

In this article, we consider the prediction of future failure times based on Type-I hybrid censored samples. Point predictors and prediction intervals using different procedures are discussed for a general model. The exponential and Rayleigh distributions are used as illustrative examples to show the most simplified forms of the so obtained predictors as well as prediction intervals. Intensive simulation study and a real life dataset are presented to illustrate our findings and results.

In survival, reliability and medical studies, it is natural to have experience with several situations pertaining to testing, cost or money constraints where the removal of units prior to failure is preplanned. In this context, we consider the inference problem including estimation and prediction for power Lindley distribution under the progressively type-II censored sample data. For the estimation purposes and other reliability characteristics maximum likelihood and Bayes approaches for estimating the model parameters are considered in this paper. Confidence intervals of the parameters and the corresponding average lengths and coverage probabilities are developed based on maximum likelihood and Bayes techniques. The Gibbs and Metropolis samplers are used to predict the life lengths of the removed units in multiple stages of the progressively censored sample. Monte Carlo simulations are performed to compare different methods and one real data set is analyzed for illustrative purposes.