We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

eess.SY

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Electrical Engineering and Systems Science > Systems and Control

Title: Few-Shot Scenario Testing for Autonomous Vehicles Based on Neighborhood Coverage and Similarity

Abstract: Testing and evaluating the safety performance of autonomous vehicles (AVs) is essential before the large-scale deployment. Practically, the number of testing scenarios permissible for a specific AV is severely limited by tight constraints on testing budgets and time. With the restrictions imposed by strictly restricted numbers of tests, existing testing methods often lead to significant uncertainty or difficulty to quantifying evaluation results. In this paper, we formulate this problem for the first time the "few-shot testing" (FST) problem and propose a systematic framework to address this challenge. To alleviate the considerable uncertainty inherent in a small testing scenario set, we frame the FST problem as an optimization problem and search for the testing scenario set based on neighborhood coverage and similarity. Specifically, under the guidance of better generalization ability of the testing scenario set on AVs, we dynamically adjust this set and the contribution of each testing scenario to the evaluation result based on coverage, leveraging the prior information of surrogate models (SMs). With certain hypotheses on SMs, a theoretical upper bound of evaluation error is established to verify the sufficiency of evaluation accuracy within the given limited number of tests. The experiment results on cut-in scenarios demonstrate a notable reduction in evaluation error and variance of our method compared to conventional testing methods, especially for situations with a strict limit on the number of scenarios.
Subjects: Systems and Control (eess.SY); Machine Learning (cs.LG); Robotics (cs.RO); Software Engineering (cs.SE)
Cite as: arXiv:2402.01795 [eess.SY]
  (or arXiv:2402.01795v2 [eess.SY] for this version)

Submission history

From: Shu Li [view email]
[v1] Fri, 2 Feb 2024 04:47:14 GMT (184kb,D)
[v2] Tue, 23 Apr 2024 03:07:18 GMT (265kb,D)

Link back to: arXiv, form interface, contact.