test
26262
SOTIF
robustness against weather conditions
[3] camera-based object detection robustness against simulated "faults" of camera noise, delay, etc. (evaluated only against one fault)
inputs / outputs (black-box)
[4] STPA guided fault injection (fault type - simulated faults - communication, unavailable, corruption)
[4] STPA guided fault injection (fault type - simulated effects of weather conditions on sensors)
ML component (white-box)
SOTIF: scenario space coverage, finding critical scenarios / test cases
system level
object detection level
[15] synthesize realistic driving scenes, different weather conditions & test inconsistencies using metamorphic testing
[16] automatically identify missed detections for camera-based object detection
[18] test case generation for system / environment critical aspects (at system level)
[19] automatic generation of test cases - different environmental conditions (weather mainly), maximize neuron coverage in test cases
[14] camera-based lane detection robustness against simulated rain
[17] falsification of formal requirements, combinatorial testing (at system level)
[11] combinatorial testing based on a domain model (at object detection level)
[1] and [2], tensorflow related but goal is to do fault injection at ML operator level + identify safety critical bits for a specific application like object detection
[5] bayesian network representation of architecture + safety constraints to guide fault injection