1
00:00:03,600 --> 00:00:09,870
So in summary, the feel that can be applied to non-linear systems and we can do this when or if it

2
00:00:09,870 --> 00:00:15,630
is possible to make appropriate assumptions or approximations which allow the nonlinear system to be

3
00:00:15,630 --> 00:00:22,230
expressed in a linear form, the increased error in the modeling process will most likely be needed

4
00:00:22,230 --> 00:00:25,870
to be accounted for by artificially increasing the process model noise.

5
00:00:26,460 --> 00:00:31,080
This is going to be a trade off with performance because as we increase the process, more noise is

6
00:00:31,080 --> 00:00:36,180
going to rely less on the prediction model, which means the field will be taking less prediction action

7
00:00:36,180 --> 00:00:38,680
and relying more on the measurements itself.

8
00:00:39,570 --> 00:00:44,550
The filter will also be needed to be tested and manually tuned to ensure the best performance.

9
00:00:45,120 --> 00:00:48,570
There's also a possibility that the field will diverge from the truth.

10
00:00:48,570 --> 00:00:52,950
And this is because of the numerical approximations that we made inside the modelling.

11
00:00:53,380 --> 00:00:58,080
If there's too much error in the model where the model differs from the actual model in real life,

12
00:00:58,440 --> 00:01:02,550
then these errors can build up and compound and cause the field of state to diverge.