只有当雅可比矩阵存在时，才能应用线性化。但是，情况并非总是如此。一些系统包含不连续性（例如，过程模型可能是跳跃线性的[14]，其中参数可以突然变化[1]，或者传感器可能返回高度量化的传感器测量值[15]），其他系统具有奇点（例如，透视投影方程[16]），而在其他系统中，状态本身本质上是离散的（例如，用于预测驾驶飞机规避行为的基于规则的系统[17]）。
Linearization can be applied only if the Jacobian matrix exists. However, this
is not always the case. Some systems contain discontinuities (for example, the
process model might be jump-linear [14], in which the parameters can change
abruptly, or the sensor might return highly quantized sensor measurements
[15]), others have singularities (for example, perspective projection equations
[16]), and in others the states themselves are inherently discrete (e.g., a
rule-based system for predicting the evasive behavior of a piloted aircraft
[17]).
[1]. Julier, S.J. and J.K. Uhlmann. Unscented filtering and nonlinear estimation. in Sequential State Estimation: From Kalman Filters to Particles Filters. 2004. Institute of Electrical and Electronics Engineers Inc.
感谢 https://www.cnblogs.com/sunny99/ sumoier对本文的帮助！
标签：linear,17,models,sensor,jump,others,Markov,might From： https://www.cnblogs.com/Yi-920259020/p/16924889.html