1 00:00:03,420 --> 00:00:09,210 So now let's go into the details and framework for the capstone project for the two day vehicle. 2 00:00:10,260 --> 00:00:16,620 So the goal, again, is to estimate the precision, velocity and orientation of a moving vehicle based 3 00:00:16,620 --> 00:00:22,140 on Leida like measurements to known landmarks and features, GPS measurements, gyroscope measurements 4 00:00:22,410 --> 00:00:24,950 and known ideal simulate the conditions. 5 00:00:25,500 --> 00:00:30,710 The problem consists of assuming that the vehicle travels at a constant speed in the 2D. 6 00:00:30,730 --> 00:00:33,180 Explain in the direction it is facing. 7 00:00:33,720 --> 00:00:39,780 Let the inputs be a turn right of the vehicle from a box gyroscope and assume that the acceleration 8 00:00:39,780 --> 00:00:41,100 is a random variable. 9 00:00:41,130 --> 00:00:43,770 So this forms a process model that we want to implement. 10 00:00:43,770 --> 00:00:44,490 In the future. 11 00:00:45,780 --> 00:00:51,330 We will assume that we get precision measurements of the vehicle location, i.e. such as from GPS that 12 00:00:51,330 --> 00:00:52,830 can have sensor faults in them. 13 00:00:52,860 --> 00:00:54,700 So this is going to be the first sensor model. 14 00:00:56,010 --> 00:01:01,470 We are also going to assume that we get a number of range and relative bearing measurements of known 15 00:01:01,470 --> 00:01:08,100 landmark locations, i.e. from a lighted like measurement with and without matching data associations. 16 00:01:08,340 --> 00:01:12,750 So this is the second sensor model that we want to use in inside this field of. 17 00:01:14,550 --> 00:01:20,730 So the presence of the filter is, again, we have a vehicle that has a 2D position, a position X and 18 00:01:20,730 --> 00:01:23,970 Y, it has a velocity and it has a heading. 19 00:01:23,970 --> 00:01:26,720 Now, the velocity is in the direction of heading. 20 00:01:26,970 --> 00:01:29,860 It can be positive or negative or it could be stationary. 21 00:01:30,480 --> 00:01:32,310 We also have a gyroscope on board. 22 00:01:32,310 --> 00:01:34,410 So we get your right information. 23 00:01:34,410 --> 00:01:36,840 We get the side from the gyroscope. 24 00:01:37,020 --> 00:01:39,210 But the gyroscope is also biased. 25 00:01:39,270 --> 00:01:44,040 So we have to now take into account the spies inside the process model and compensate for it. 26 00:01:44,850 --> 00:01:52,650 The acceleration can be model as a random variable with a zero main and a noise variance of Sigma V 27 00:01:52,650 --> 00:01:53,160 squared. 28 00:01:54,510 --> 00:02:00,720 The first since Emoto, the GPS measurements can just be a simple position in the X and Y that is turned 29 00:02:00,720 --> 00:02:01,650 by the sensor. 30 00:02:02,010 --> 00:02:07,470 However, this sensor model is also going to have GPS tonight, areas to areas inside the simulation 31 00:02:07,470 --> 00:02:12,720 environment where we get zero measurements and we can also have a random amount of faulty measurements 32 00:02:12,720 --> 00:02:14,690 inside the GPS measurements. 33 00:02:15,690 --> 00:02:21,720 The second sensor is the lie to lock sensor measurement, so the lighter again produces range and relative 34 00:02:21,720 --> 00:02:24,150 bearing measurements to landmark locations. 35 00:02:24,570 --> 00:02:31,710 However, now we also have to deal with no landmark identification and data association so we can detect 36 00:02:31,710 --> 00:02:34,320 the range and relative bearing to a landmark. 37 00:02:34,500 --> 00:02:37,240 But we might not know which landmark we're actually seeing. 38 00:02:37,260 --> 00:02:38,970 There's no data association with that. 39 00:02:38,970 --> 00:02:43,110 So we're going to have to come up with a process to do this data association for us. 40 00:02:44,970 --> 00:02:50,760 So now let's have a look at the simulation profiles of profiles nine and zero that we're going to be 41 00:02:50,760 --> 00:02:53,040 using for the capstone project. 42 00:02:53,940 --> 00:02:58,020 So in this run here, you can see the common thread is actually running. 43 00:02:58,020 --> 00:03:03,180 We have the uncertainty ellipse and this is the uncertainty ellipse and common field of vision. 44 00:03:03,180 --> 00:03:09,330 If we're just running the unscented common filter that we produced in the last example so we can see 45 00:03:09,330 --> 00:03:14,640 that the vehicle starts at a non-zero initial condition and in fact starts moving in reverse. 46 00:03:14,670 --> 00:03:18,900 So this is going to be the first set of challenges that you encounter when you're using this capstone 47 00:03:18,900 --> 00:03:21,660 project profile once a vehicle starts moving. 48 00:03:21,690 --> 00:03:27,060 We then start to get the Leida measurements and then we enter this orange circle sign here. 49 00:03:27,090 --> 00:03:29,670 So this is when we enter the GPS denied region. 50 00:03:29,670 --> 00:03:34,160 So you can see that we stop getting any GPS measurements inside the circle. 51 00:03:34,620 --> 00:03:39,980 So we're going to be totally relying on the gyroscope and the leider information during this region. 52 00:03:40,350 --> 00:03:44,490 So you can see that the estimated position from the unscented coming forward is moving all over the 53 00:03:44,490 --> 00:03:47,240 place because we don't have valid information. 54 00:03:47,880 --> 00:03:53,640 Once the vehicle leaves the GPS not there, you can see that the GPS measurements come back and then 55 00:03:53,640 --> 00:04:00,090 we go into reverse for a slow period, moving in the reverse direction before the vehicle starts to 56 00:04:00,090 --> 00:04:02,160 accelerate and then starts moving forward. 57 00:04:02,490 --> 00:04:08,400 Towards the end of the profile, we enter back into the GPS to concern before turning back and moving 58 00:04:08,400 --> 00:04:11,520 closer towards to where we initialize the fuel from. 59 00:04:12,000 --> 00:04:17,610 So you can see from the whole profile, the estimated error, the main square there for the position, 60 00:04:17,790 --> 00:04:21,000 the velocity and the heading overall quite large. 61 00:04:21,000 --> 00:04:23,960 The field, it does not perform very well in this situation. 62 00:04:24,300 --> 00:04:29,910 So this is a challenge to try to get the filter to produce as small as estimates as possible over this 63 00:04:29,910 --> 00:04:33,180 profile and the profile without the data association. 64 00:04:33,780 --> 00:04:38,220 Your filter solution should also work for all the other profiles that you can access. 65 00:04:38,220 --> 00:04:41,070 It should be seamlessly working with everything.