Testing DNN-based Autonomous Driving Systems under Critical Environmental Conditions

Authors: Zhong Li, Minxue Pan, Tian Zhang, Xuandong Li

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Large-scale experiments show that TACTIC can effectively identify critical environmental conditions and produce realistic testing images, and meanwhile, reveal more erroneous behaviours compared to existing approaches.
Researcher Affiliation Academia 1State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China 2Department of Computer Science and Technology, Nanjing University, Nanjing, China 3Software Institute, Nanjing University, Nanjing, China.
Pseudocode No The paper describes the TACTIC approach in detail but does not provide any structured pseudocode or algorithm blocks.
Open Source Code Yes Our implementation of TACTIC and all the experimental data are publicly available1 to facilitate future studies. 1https://github.com/SEG-DENSE/TACTIC
Open Datasets Yes We conduct experiments on the Udacity dataset (Udacity, 2016).
Dataset Splits No The paper mentions using the Udacity dataset and its testing portion, but it does not specify explicit training/validation splits or their sizes for model training, nor does it refer to standard predefined splits for the purpose of reproduction.
Hardware Specification No The paper refers to an 'experimental environment' and mentions 'time cost', but it does not provide specific hardware details such as GPU/CPU models, memory, or cloud instance types used for the experiments.
Software Dependencies No The paper mentions the use of MUNIT and DNN-based ADSs but does not provide specific version numbers for any software dependencies, libraries, or frameworks used in the implementation.
Experiment Setup Yes We set the number k of KMNC to 1000, consistent with Deep Gauge (Ma et al., 2018). For the number of detected erroneous behaviours, we present the number of erroneous behaviours detected under four different error bounds (c.f. Section 3.2.2)... In this work, the stopping condition is set conservatively to that the fitness value has no improvement in 100 successive iterations. In this work, we set the number [of critical conditions] to be four...