Research map / Probabilistic and scientific modeling / Molecules and proteins
Monitoring of Perception Systems: Deterministic, Probabilistic, and Learning-Based Fault Detection and Identification (Abstract Reprint)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Formalizes runtime fault detection and identification for perception systems via diagnostic graphs, with deterministic (optimization), probabilistic (factor graph MAP), and GNN-based inference, plus diagnosability guarantees and LGSVL/Apollo experiments.
Problem
Runtime monitoring of perception systems lacks a system-level framework for detecting and identifying which modules or outputs have failed, given heterogeneous, possibly asynchronous diagnostic tests.
Approach
Model diagnostic tests and a priori constraints as a bipartite diagnostic graph over failure-mode variables, then perform fault identification via integer programming (deterministic), factor-graph MAP inference (probabilistic), or a graph neural network, with diagnosability results from multiprocessor fault diagnosis theory.
Key results
- Some proposed monitors outperform common baselines in fault identification accuracy on simulated Apollo failure scenarios.
- Monitor runtime is typically below five milliseconds, a negligible overhead.
- Monitors provide enough notice to stop the vehicle before accidents in realistic scenarios.
Limitations
- Fault identification relies on the designer knowing the failure modes in advance
- Integer programming is NP-complete, though problems are small in practice
- Diagnosability guarantees hold only under technical assumptions such as weak-OR tests with scope size 2
Experimental setup
Datasets: LGSVL simulator-generated driving scenarios
Benchmarks: LGSVL simulator scenarios with Apollo Auto
Metrics: accuracy runtime
Baselines: Baidu Apollo built-in checks
Builds on: PMC model diagnosability structured SVM GCN GCNII GraphSAGE
Closest papers by meaning
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.86)
- DME-Driver: Integrating Human Decision Logic and 3D Scene Perception in Autonomous Driving (AAAI 2025 · similarity 0.84)
- DriveHierarchy: A Benchmark for Diagnosing VLM Driving Capabilities from Open-Loop Understanding to Closed-Loop Execution (NeurIPS 2026 · similarity 0.84)
- Continuously Learning, Adapting, and Improving: A Dual-Process Approach to Autonomous Driving (NeurIPS 2024 · similarity 0.84)
- SafeAuto: Knowledge-Enhanced Safe Autonomous Driving with Multimodal Foundation Models (ICML 2025 · similarity 0.84)
- Hybrid-Driving: An Autonomous Driving Decision Framework Integrating Large Language Models, Knowledge Graphs and Driving Rules (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.