Probabilistic Risk Assessment of an Obstacle Detection System for GoA 4 Freight ...
summary
Five-step quantitative procedure estimates false negative hazard for a GoA4 freight obstacle detector by composing a fault tree with parameter values from CTMC with parametric model checking, using Monte Carlo and statistical tests to bound perception errors and check channel independence with a chi square test. Neural network residual risk is upper-bounded via a coupon collector for NN coverage over image equivalence classes. Combining modules into a two dual-channel modules plus voter architecture reaches the cited tolerable hazard rate of 1e-7 under stated assumptions.
AI-written summary · about
chapters (17)
- 0:00Probabilistic risk assessment for freight trains
- 1:07GoA 4 and false negatives in obstacle detection
- 2:11Fault-tolerant sensor architecture and voting
- 3:16Risk assessment in autonomous freight trains
- 4:00Fault tree analysis and Markov chains
- 5:07Risk assessment procedure and certification limits
- 6:32Hazard probability and tolerable rates
- 8:17Fault tree construction and error probabilities
- 10:22Failure modes in two-channel sensors
- 12:13Perception complexity and neural networks
- 13:20Neural network misclassification probabilities
- 17:55Coupon collector problem and equivalence classes
- 19:45Mathematical bounds and independence checks
- 20:52Encoding basic events in model
- 21:56Combining modules for certification
- 22:39GoA 4 role and practical takeaways
- 23:03Q&A on method limits and real-world use
related talks
13mWIP: Learning Adversarial Attacks on Adaptive Traffic Signal Control Systems...
49mThe Math of Reliability
27mSymbolic Reachability Analysis of Distributed Systems using Narrowing and Heuristic Search
18mEfficient Exposure of Partial Failure Bugs in Distributed Systems with Inferred...
21mDependency Safety for Distributed Systems
33mPruning Redundancy in Answer Set Optimization Applied to Preventive Maintenance ...
39mBroadcasting on Directed Acyclic Graphs
16mAI Investigations Are a Distributed System Problem