EvoEye: Self-Evolving Runtime Monitoring for Autonomous Driving Systems

2016-01-01 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Expert, extended

Summary

EvoEye is a self-evolving runtime monitoring framework designed for autonomous driving systems (ADSs) to detect impending hazards. It addresses limitations of fixed-capability monitors by combining FusionMonitor, a learning-based component that models cross-module temporal interactions for collision prediction, with BlindSpotEvolver, which identifies current monitor errors and generates informative executions for updates. Evaluated on Baidu Apollo with CARLA in representative highway and urban scenarios, FusionMonitor improved frame-level Recall by up to 37.8 percentage points at a 0.05 false positive rate, exhibiting 2.49 ms latency and 2.8–4.2 seconds median warning time. BlindSpotEvolver further enhanced performance by up to 13.2 F1 points on previously missed unsafe contexts compared to other sampling methods.

Key takeaway

For Machine Learning Engineers developing safety-critical autonomous driving systems, you should consider implementing self-evolving runtime monitors like EvoEye. This approach ensures your system's hazard detection capabilities continuously adapt and improve by actively seeking out and learning from scenarios that expose current monitoring blind spots. Prioritize integrating cross-module runtime signals and a feedback-driven scenario acquisition strategy to maximize monitoring effectiveness and reduce false alarms.

Key insights

Self-evolving runtime monitors for ADSs improve hazard detection by iteratively learning from targeted, error-revealing scenario executions.

Principles

• System-level risk prediction requires modeling cross-module temporal interactions.
• Monitor-guided data acquisition efficiently targets current detection weaknesses.
• Density-aware mutation balances local exploitation and global exploration in scenario search.

Method

EvoEye trains an initial FusionMonitor, then BlindSpotEvolver iteratively acquires new scenarios using monitor prediction errors and density-aware mutation, updating the monitor with new records.

In practice

• Integrate module-level runtime signals for comprehensive ADS risk assessment.
• Prioritize scenario generation based on current monitor's prediction errors.

Topics

• Autonomous Driving Systems
• Runtime Monitoring
• Self-Evolutionary Algorithms
• Collision Prediction
• Baidu Apollo
• CARLA Simulator
• Machine Learning Safety

Code references

• autowarefoundation/autoware

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• What Probing Reveals about Autonomous Driving: Linking Internal Prediction Errors to Ego Planning — Autonomous driving policies often lack timely surrounding-vehicle predictions during near-collisions, impacting ego planning despite good overall performance.
• Learning collision risk proactively from naturalistic driving data at scale — GSSM proactively learns collision risk from naturalistic driving data, outperforming baselines in accuracy and timeliness.
• EvoTest: Evolutionary Test-Time Learning for Self-Improving Agentic Systems — EvoTest enables AI agents to self-improve at test time by evolving their entire configuration using narrative feedback.
• Metis: A Generalizable and Efficient World-Action Model for Autonomous Driving and Urban Navigation — Decoupling video generation from action prediction in WAMs, using a Mixture-of-Transformers and asymmetric attention, improves efficiency and generalization for autonomous navigation.
• Bridging Local Observation and Global Simulation in Closed-Loop Traffic Modeling — CRAFT mitigates local-to-global context mismatch in traffic simulators via self-supervised failure discovery and preference-guided alignment.
• Towards Safety-Aware Mutation Testing for Autonomous Driving Systems — SAMT shifts ADS testing from component reliability to system safety by mutating inter-module messages based on safety engineering principles.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.