Unverified50% confidenceFactExact time
Domain randomization can only cover error sources that the developer has already identified and explicitly modeled; once the real robot exhibits unmodeled factors, the policy is exposed to training blind spots.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
Unverified即便固定所有随机种子，在不同硬件或并行度下，训练结果仍可能出现被大模型放大为显著能力差异的细微偏差76% similarUnverifiedThe developer used a tool called Heretic to perform abliteration; the tool is inherently stochastic, finding a slightly different refusal direction with each run.75% similarUnverified过度域随机化会引入与真实分布相差过大的不可能样本，稀释有效训练信号，随机化范围的设定需要大量领域专家经验75% similarVerified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74% 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.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/968878API
curl https://kongchang.com/api/v1/knowledge/claims/968878MCP
get_claim(id=968878)