1
00:00:03,810 --> 00:00:10,410
Now that we have finished the unscented common update step, we can look at the results of the simulation.

2
00:00:11,460 --> 00:00:16,860
So the results between the extended common filter and the antennae coming through are very similar to

3
00:00:17,190 --> 00:00:22,590
the additional accuracy provided by the incoming filter is not overall beneficial.

4
00:00:22,590 --> 00:00:30,150
In this specific case, the system models are still fairly linear around the truth, and the first order

5
00:00:30,150 --> 00:00:35,880
Jacoby model used in the extended common filter can work appropriately for these models.

6
00:00:36,270 --> 00:00:41,900
However, will be different situations with different dynamics where this will not be the case.

7
00:00:42,420 --> 00:00:48,180
So if we look at the results we get between the extended camera filter and the antenna karmi filter,

8
00:00:48,600 --> 00:00:50,850
we get results that look a bit like this.

9
00:00:51,270 --> 00:00:56,730
And you can see between the antenna a filter and the external coming filter for the different profiles.

10
00:00:57,010 --> 00:01:00,660
You can see that the results are very closely match.

11
00:01:01,150 --> 00:01:04,850
There's only sort of one result that sort of stands out by looking at them over here.

12
00:01:04,860 --> 00:01:07,420
And this is for the profile number six.

13
00:01:08,120 --> 00:01:14,970
So this is when we start with the non-zero initial condition, the initial transient period as filters

14
00:01:14,970 --> 00:01:17,880
converging is different, which is why we get different results.

15
00:01:18,210 --> 00:01:22,760
So this is not to say the unsuspended coming here is not performing as well as the echo.

16
00:01:23,070 --> 00:01:28,440
It's just that the method it goes after the first couple of updates from the initialization is going

17
00:01:28,440 --> 00:01:31,860
to be slightly different because we're using a different update step process.

18
00:01:32,610 --> 00:01:36,660
So all this error here is just due to the initial convergence.

19
00:01:37,440 --> 00:01:42,230
Once the filter converges, then it should be performed pretty much exactly same as IQF.

20
00:01:42,570 --> 00:01:46,140
So all of this is due to initial transition and the convergence.

21
00:01:46,620 --> 00:01:50,800
And we can see that if we look at the simulation results side by side.

22
00:01:51,930 --> 00:01:59,190
So now let's run the IQF and the UK at the same time and look at the initial condition, transient results.

23
00:02:00,170 --> 00:02:04,860
So you can see as the field starts to run, it jumps around a bit more on the UK.

24
00:02:05,090 --> 00:02:10,280
So this is why we get the larger error overall is just because of the initial convergence.

25
00:02:10,280 --> 00:02:17,540
It takes slightly longer for it to converge down to the truth while the akef converges a lot more quickly

26
00:02:17,780 --> 00:02:20,510
in this specific case is not going to be in general.