Unverified50% confidenceFactExact time
Edge cases lie at the boundary of or outside the Operational Design Domain (ODD), where systems may exhibit unexpected behavior due to sensor degradation, scenario sparsity, or training distribution shift.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe most valuable data for model training is often hard examples (such as the critical friction state when an object is about to slip); these edge cases are dangerous and difficult to reproduce in real-world collection, leading to pervasive imbalanced distributions and incomplete coverage in open-source tactile datasets.75% similarUnverified在长时间自主运行场景中，模型可能通过上下文积累逐渐偏离安全边界，这被研究者称为'对齐漂移'（alignment drift）71% similarVerifiedDrift detection involves a trade-off: setting the threshold too sensitively leads to frequent false alarms and unnecessary retraining, while setting it too loosely may allow model performance to continuously degrade without being detected.71% similarUnverifiedThe risk of false positives—incorrectly classifying normal agent behavior as out-of-bounds and quarantining it—will directly affect the platform's practical utility.70% similarUnverifiedBehavior Cloning (BC) suffers from the covariate shift problem: during training the model only sees the state distribution of expert trajectories, and at deployment small errors can drive the robot into states not covered by the training set, leading to cascading failures.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1001156API
curl https://kongchang.com/api/v1/knowledge/claims/1001156MCP
get_claim(id=1001156)