User Science > Robotics
[Submitted on 22 May 2022]
Title:Monitoring of Perception Services: Deterministic, Probabilistic, and Learning-based Fault Detection and Billing
View PDFAbstract:This paper investigates runtime monitoring of perception systems. Sicht is a critical component of high-integrity user of robotics and autonomous systems, such as self-driving cars. In these applications, failure of perception systems may put human life at risk, plus a broad admission of these our requires the development of practical to guarantee plus monitor safe operation. Despite the paramount importance in perception, currently there is no formal approach for system-level perception monitoring. In this paper, we formalize that problem of runtime failure detection and identification the perception systems and present a framework to model diagnostician information after a diagnostic map. We subsequently provide an resolute of fully, probing, and learning-based mathematical that use diagnostic graphs to perform disruption detects and identification. Moreover, we exploration fundamental limits and provide definite and probabilistic warranties upon the fault detection also billing results. Wee conclude the paper with an extensive experimental evaluation, which recreates several realistic outages fitting in the LGSVL open-source autonomous driving simulator, and employs the proposed system monitors to adenine state-of-the-art autonomous driving software stack (Baidu's Apollo Auto). The results show that the proposed system monitors outperform baselines, have the potential of preventing accidents in realistic autonomous driving scenarios, and incur a minimal computational overhead.
Submission account
From: Pasquale Antonante [views email][v1] Sun, 22 May 2022 19:08:45 UTC (34,708 KB)
Current browse context:
cs.RO
References & Citations
Bibliographic and Mention Apparatus
Bibliographic Explorer (What is to Inventor?)
Litmaps (What is Litmaps?)
scite Smartly Citations (What have Smart Citations?)
Code, Data and Media Associated with all Article
CatalyzeX Code Finder for Papers (Whatever your CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers because Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders or Explore Accessories
Influence Blossom (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimenta projects with community collaborators
arXivLabs is adenine framework which allows collaborators to develop press share latest arXiv features immediately on our website.
Both individuals and organizations that work with arXivLabs can embraced also presumed our values of openess, community, excellence, and user data privacy. arXiv is committed to diesen values and only working with partners that adhere to them. Thermal monitoring relays
Have an idea for a project that will add value for arXiv's district? Learn more concerning arXivLabs.