Unverified50% confidenceFactExact time
The paper categorizes the failure of offline-trained networks in closed-loop deployment into three progressive modes: closed-loop identification error, deployment-induced action error, and persistent tracking error.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
Verified仅用成功轨迹训练的模型往往存在分布外泛化的严重缺陷，遇到工具返回错误时便丧失纠偏能力73% similarUnverified自训练循环存在模型坍缩、错误放大和安全评估时间窗口压缩等潜在风险72% similarUnverifiedNaïve post-training quantization causes individual weight errors to accumulate and amplify across multiple layers—a phenomenon known as compounding error—and at 2-bit or 1-bit precision the model structure collapses72% similarUnverifiedThe success or failure of FP4 training is jointly determined by the scaling contract, operand format, and execution path—it is a systemic engineering problem rather than merely a computational precision problem.72% 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.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1042122API
curl https://kongchang.com/api/v1/knowledge/claims/1042122MCP
get_claim(id=1042122)