WEBVTT 00:00.420 --> 00:02.850 -: Hello and welcome to this tutorial. 00:02.850 --> 00:05.820 In the previous tutorial, we initialized the map. 00:05.820 --> 00:07.830 And now, time for the exciting stuff. 00:07.830 --> 00:09.900 We create the car. 00:09.900 --> 00:12.060 And we do that with a class of course. 00:12.060 --> 00:14.580 You will see that the class is very practical 00:14.580 --> 00:18.360 to create some things that have a lot of properties, 00:18.360 --> 00:19.560 because as you can see, 00:19.560 --> 00:22.470 not only I'm defining some variables for my car, 00:22.470 --> 00:24.570 but also some functions, 00:24.570 --> 00:27.150 which of course is the function that will make the car 00:27.150 --> 00:30.660 move to the left, to the right, or going straight. 00:30.660 --> 00:32.400 So we have a couple of variables 00:32.400 --> 00:35.190 that are important to describe the environment. 00:35.190 --> 00:36.870 We have, for example, the angle, 00:36.870 --> 00:39.120 which is the angle between the X-axis 00:39.120 --> 00:42.600 and the axis of the direction of the car. 00:42.600 --> 00:46.230 Then we have the rotation, which is its last rotation, 00:46.230 --> 00:48.930 which remembers either zero degree, 00:48.930 --> 00:51.540 20 degrees, or minus 20 degrees. 00:51.540 --> 00:53.430 Then we have the velocity, 00:53.430 --> 00:55.890 the X-coordinate of the velocity vector 00:55.890 --> 00:58.440 and the Y-coordinate of the velocity vector, 00:58.440 --> 00:59.730 and then the vector 00:59.730 --> 01:03.390 of coordinates velocity X and velocity Y. 01:03.390 --> 01:05.790 Then we have the sensors and the signals, 01:05.790 --> 01:07.380 and that's very important. 01:07.380 --> 01:10.470 The car that we're making will have three sensors, 01:10.470 --> 01:13.470 sensor one, sensor two, and sensor three. 01:13.470 --> 01:15.870 Sensor one will be detecting 01:15.870 --> 01:18.420 if there is any sand in front of the car. 01:18.420 --> 01:21.540 Then sensor two is the sensor that will detect 01:21.540 --> 01:24.420 if there is any sand at the left of the car. 01:24.420 --> 01:27.000 And sensor three is the sensor that will detect 01:27.000 --> 01:29.910 if there is any sand at the right of the car. 01:29.910 --> 01:33.930 And then from these three sensors, we get the signals, 01:33.930 --> 01:37.290 that is the signals received by each of the sensors. 01:37.290 --> 01:40.350 So signal one is the signal received by sensor one. 01:40.350 --> 01:43.110 Signal two is the signal received by sensor two, 01:43.110 --> 01:46.410 and signal three is the signal received by sensor three. 01:46.410 --> 01:47.760 And so how does it work? 01:47.760 --> 01:51.900 Signal one is the density of sand around sensor one, 01:51.900 --> 01:54.660 signal two is the density of sand around sensor two, 01:54.660 --> 01:58.260 and signal three is the density of sand around sensor three. 01:58.260 --> 02:00.660 And how do we compute this density of sand? 02:00.660 --> 02:01.710 Well, that's very simple. 02:01.710 --> 02:06.120 We take some big squares around each of the sensors. 02:06.120 --> 02:09.030 These are actually squares of 200 by 200. 02:09.030 --> 02:10.500 And for each of the squares, 02:10.500 --> 02:13.170 we divide the number of ones in the square 02:13.170 --> 02:15.360 by the total number of cells in the square. 02:15.360 --> 02:18.030 That is, 20 times 20 equals 400. 02:18.030 --> 02:20.220 And that gives us the density of sand 02:20.220 --> 02:22.590 because the ones correspond to the sand. 02:22.590 --> 02:24.390 We do this for each sensor, 02:24.390 --> 02:26.580 and that gives us the density of sand 02:26.580 --> 02:29.610 around each sensor, that is, the signals. 02:29.610 --> 02:31.950 All right, so now we have everything to detect the sand 02:31.950 --> 02:35.460 and then we have the move function. 02:35.460 --> 02:38.100 And of course, the move function is what will allow the car 02:38.100 --> 02:41.880 to go to the left, going straight, or going to the right. 02:41.880 --> 02:43.350 So let's go through it quickly. 02:43.350 --> 02:46.890 We have here the update of the position of the car 02:46.890 --> 02:50.700 with its last position, which is self.pos here, 02:50.700 --> 02:52.530 and the velocity vector. 02:52.530 --> 02:54.900 So thanks to the velocity vector, 02:54.900 --> 02:56.520 the position will be updated 02:56.520 --> 02:58.890 in the direction of the velocity vector. 02:58.890 --> 03:01.200 Then we get the rotation, 03:01.200 --> 03:05.280 which we will get further down in the code right here. 03:05.280 --> 03:08.040 Rotation equals action-to-rotation action. 03:08.040 --> 03:09.480 Here we will select the action 03:09.480 --> 03:11.400 and then getting the rotation. 03:11.400 --> 03:15.600 And so, this self-rotation equals rotation here 03:15.600 --> 03:17.580 is this rotation that we get 03:17.580 --> 03:19.740 to know how we need to rotate the car 03:19.740 --> 03:22.410 that is going to the left or to the right. 03:22.410 --> 03:24.540 Then we update the angle, 03:24.540 --> 03:27.240 which I remind is the angle between the X-axis 03:27.240 --> 03:29.580 and the axis of the direction of the car. 03:29.580 --> 03:31.350 And then once the car has moved, 03:31.350 --> 03:34.650 then we have to update the sensors and the signal. 03:34.650 --> 03:37.560 Because of course, when the car has just rotated, 03:37.560 --> 03:39.900 well, the sensors have rotated as well, 03:39.900 --> 03:41.910 and therefore we need to rotate them 03:41.910 --> 03:43.590 by using the rotate function 03:43.590 --> 03:45.720 and to which we add the new position. 03:45.720 --> 03:49.080 And why do we have this vector of 30,0? 03:49.080 --> 03:51.930 Well, that's simply because 30 is the distance 03:51.930 --> 03:54.090 between the car and the sensor. 03:54.090 --> 03:56.220 You know it is the distance between the car 03:56.220 --> 03:59.160 and what the car detects. 03:59.160 --> 04:01.890 And then once the sensors are updated, 04:01.890 --> 04:04.230 well, then it's time to update the signals. 04:04.230 --> 04:05.910 And so here we do exactly what I explained 04:05.910 --> 04:07.200 to compute the signals. 04:07.200 --> 04:09.570 We get the X-coordinate of our sensor. 04:09.570 --> 04:13.350 Then we take all the cells from minus 10 to plus 10. 04:13.350 --> 04:15.420 Then we do the same for the Y-coordinate, 04:15.420 --> 04:19.080 taking all the cells from minus 10 to plus 10. 04:19.080 --> 04:23.250 So therefore, we get the square of 20 by 20 pixels 04:23.250 --> 04:24.990 surrounding the sensor. 04:24.990 --> 04:28.260 And inside this square, we sum all the ones. 04:28.260 --> 04:29.880 So basically, we sum all the cells, 04:29.880 --> 04:33.120 because the cells contain either zero or one. 04:33.120 --> 04:33.953 And since 04:33.953 --> 04:38.160 in a 20-by-20 square, there is 20 times 20 equals 400 cells, 04:38.160 --> 04:40.200 well, we divided by 400 04:40.200 --> 04:43.500 to get the density of ones inside the square. 04:43.500 --> 04:45.720 And that's how we get the signal 04:45.720 --> 04:48.570 of the density of sand around the sensor. 04:48.570 --> 04:51.060 And we do the same for the second sensor 04:51.060 --> 04:52.350 and the third sensor 04:52.350 --> 04:55.110 to get the second signal and the third signal. 04:55.110 --> 04:56.940 Okay, so that's to detect the sand. 04:56.940 --> 05:00.210 And then the three lines of code here are very important. 05:00.210 --> 05:03.270 It's another bad reward that we want to give to our car 05:03.270 --> 05:06.450 when it is reaching one of the edges of the map. 05:06.450 --> 05:09.420 We don't want the car to rush into some walls 05:09.420 --> 05:11.220 and therefore we want to penalize it, 05:11.220 --> 05:14.610 to punish it when it's getting too close to a wall. 05:14.610 --> 05:16.590 And therefore, that's what we do here. 05:16.590 --> 05:20.460 If the first sensor is larger than longer minus 10, 05:20.460 --> 05:23.160 that is, larger than here, 05:23.160 --> 05:25.200 because longer is this distance here. 05:25.200 --> 05:27.570 So longer minus 10 is right here. 05:27.570 --> 05:31.080 So if sensor one X larger than longer minus 10 05:31.080 --> 05:33.540 concerns all the points that are here, 05:33.540 --> 05:35.700 that is, if the car is getting closer 05:35.700 --> 05:37.740 to the right edge of the map. 05:37.740 --> 05:42.480 Or if cell sensor one X lower than 10, that's right here, 05:42.480 --> 05:45.900 if the car is getting closer to the left edge of the map. 05:45.900 --> 05:49.680 Or if sensor y is larger than larger minus 10, 05:49.680 --> 05:52.140 that's the upper edge of the map. 05:52.140 --> 05:55.680 And or if self-sensor y is lower than 10, 05:55.680 --> 05:58.680 that is, the lower edge of the map. 05:58.680 --> 06:01.590 And so if the sensor one is reaching 06:01.590 --> 06:03.180 any of these four edges, 06:03.180 --> 06:06.480 well, we will put the signal of the sensor, 06:06.480 --> 06:08.790 signal one is the signal of sensor one, 06:08.790 --> 06:10.500 we will set it to be one. 06:10.500 --> 06:11.610 And what does that mean? 06:11.610 --> 06:15.540 That means full sand, like the full density of sand. 06:15.540 --> 06:17.280 It's like the worst sand you could get. 06:17.280 --> 06:19.650 There is so much sand that it's going to stop your car. 06:19.650 --> 06:21.120 So signal will be one, 06:21.120 --> 06:24.450 and therefore the car will get a terribly bad reward. 06:24.450 --> 06:26.550 All right, and then we do the same for signal two 06:26.550 --> 06:29.553 and signal three from sensor two and sensor three. 06:30.390 --> 06:33.030 All right. And then we create the Game class. 06:33.030 --> 06:35.640 So that's basically the class to create the game 06:35.640 --> 06:38.460 because so far we have only created the car. 06:38.460 --> 06:40.770 And now of course, we have to create the map. 06:40.770 --> 06:42.360 We have to create the game itself. 06:42.360 --> 06:44.520 So we will not be playing the game. 06:44.520 --> 06:46.920 It's our AI that will be playing the game. 06:46.920 --> 06:48.900 And the game is actually to avoid the obstacles 06:48.900 --> 06:52.500 and to go from the airport to downtown and vice versa. 06:52.500 --> 06:54.360 So in this Game class, 06:54.360 --> 06:57.570 we need to create some objects, like the car. 06:57.570 --> 07:00.570 Then we need to define the update function. 07:00.570 --> 07:01.770 That is the most important 07:01.770 --> 07:03.870 and actually we will focus on that right now, 07:03.870 --> 07:05.730 because that's in this update function 07:05.730 --> 07:09.570 that we will select the action that the car has to do 07:09.570 --> 07:12.330 at each time to accomplish its goal. 07:12.330 --> 07:16.830 And this action is exactly the output of our neural network, 07:16.830 --> 07:18.960 the neural network that will be at the heart 07:18.960 --> 07:20.760 of our artificial intelligence. 07:20.760 --> 07:24.240 And so this action is returned by the brain of the car, 07:24.240 --> 07:27.420 which I remind is the object of our DQN class 07:27.420 --> 07:29.880 that we'll be making in our AI file. 07:29.880 --> 07:33.420 And this object has a method that is called update, 07:33.420 --> 07:37.590 and it takes as input the last reward and the last signal. 07:37.590 --> 07:40.140 So the last reward is of course the last reward obtained 07:40.140 --> 07:41.220 by the car. 07:41.220 --> 07:44.430 And the last signal is of course the last signal 07:44.430 --> 07:45.630 of the three sensors. 07:45.630 --> 07:47.580 Signal one from sensor one. 07:47.580 --> 07:49.350 Signal two from sensor two. 07:49.350 --> 07:51.420 Signal three from sensor three. 07:51.420 --> 07:53.490 But then I'm adding two other inputs, 07:53.490 --> 07:55.200 which is the orientation of the car 07:55.200 --> 07:56.880 with respect to the goal. 07:56.880 --> 07:59.490 So for example, if the car is heading towards the goal, 07:59.490 --> 08:01.680 then the orientation will be equal to zero. 08:01.680 --> 08:03.840 If it goes slightly to the right, 08:03.840 --> 08:06.900 then the orientation will be close to 45 degrees. 08:06.900 --> 08:08.520 And if it goes slightly to the left, 08:08.520 --> 08:11.880 the orientation will be close to minus 45 degrees. 08:11.880 --> 08:15.390 So that's the fourth input of our input state. 08:15.390 --> 08:18.990 And then there's a last input, which is minus orientation. 08:18.990 --> 08:22.590 So usually, the inputs of a neural network are independent. 08:22.590 --> 08:24.750 There is no multicollinearity. 08:24.750 --> 08:26.340 But it doesn't really matter if we add this 08:26.340 --> 08:28.470 because the neural network will just fix that 08:28.470 --> 08:29.460 with the weights. 08:29.460 --> 08:33.570 But still, I noticed that by adding this minus orientation, 08:33.570 --> 08:34.860 well, that allows the car, 08:34.860 --> 08:38.190 the training of the car, to stabilize the exploration. 08:38.190 --> 08:41.370 We're doing this so that the AI doesn't always explore 08:41.370 --> 08:42.780 in the same direction. 08:42.780 --> 08:44.400 By adding this minus orientation, 08:44.400 --> 08:47.820 we make sure that it explores in both directions, 08:47.820 --> 08:48.900 right or left. 08:48.900 --> 08:49.800 And so this, 08:49.800 --> 08:53.670 the three signals plus the orientation and minus orientation 08:53.670 --> 08:57.060 are the five inputs of our encoded vector, 08:57.060 --> 08:59.040 which will go into the network. 08:59.040 --> 09:02.880 That's our input vector that will go into the network. 09:02.880 --> 09:04.920 And after it goes into the network, 09:04.920 --> 09:07.830 well, the network returns the output, 09:07.830 --> 09:10.740 which is the action to play at each time. 09:10.740 --> 09:13.560 And the output is returned by this update function 09:13.560 --> 09:15.150 that contents the network itself 09:15.150 --> 09:16.770 and the output of the network. 09:16.770 --> 09:19.770 And therefore, that's why we have to input the last signal, 09:19.770 --> 09:21.240 that is, the input state, 09:21.240 --> 09:22.530 and also the last reward 09:22.530 --> 09:26.910 because the action to play also depends on the last reward. 09:26.910 --> 09:30.330 All right. And then we update the mean score of the rewards. 09:30.330 --> 09:32.370 We update the rotation. 09:32.370 --> 09:35.220 We use the move function to rotate the car 09:35.220 --> 09:37.440 according to the action that was selected. 09:37.440 --> 09:40.860 We update the distance of the car to the goal. 09:40.860 --> 09:43.560 And we update the positions of the sensors. 09:43.560 --> 09:46.350 Ball one, ball two, and ball three correspond to the balls 09:46.350 --> 09:48.480 that will represent the sensors on the map. 09:48.480 --> 09:50.520 You will see that very quickly. 09:50.520 --> 09:53.040 And then here, that part is very important 09:53.040 --> 09:55.020 because that's where we penalize the car 09:55.020 --> 09:57.000 if it goes onto some sand. 09:57.000 --> 09:59.310 Because as you can see, this means 09:59.310 --> 10:04.140 if the car is onto some sand, well, it will be slowed down. 10:04.140 --> 10:06.810 So that's where we reduce its velocity. 10:06.810 --> 10:10.110 Its velocity is usually six, as you can see here. 10:10.110 --> 10:12.690 And if it goes onto some sand, it will be one. 10:12.690 --> 10:14.400 So it will be slowed down to one. 10:14.400 --> 10:16.170 You'll see how the car will be slowed down 10:16.170 --> 10:18.060 once it goes onto some sand. 10:18.060 --> 10:19.200 So it is slowed down, 10:19.200 --> 10:21.840 and besides, it gets a bad reward. 10:21.840 --> 10:23.910 It gets a minus one reward. 10:23.910 --> 10:26.100 And that's actually the worst reward you could get. 10:26.100 --> 10:29.520 The best reward is one, the worst reward is minus one, 10:29.520 --> 10:32.550 and the reward is between minus one and plus one. 10:32.550 --> 10:35.970 And then otherwise, if the car isn't onto some sand, 10:35.970 --> 10:39.540 well, it keeps its usual speed, speed of six. 10:39.540 --> 10:41.040 And then we add something else. 10:41.040 --> 10:43.170 If it's getting closer to the goal, 10:43.170 --> 10:45.870 then it will get a slightly positive reward. 10:45.870 --> 10:48.150 And if it's getting further away from the goal, 10:48.150 --> 10:52.023 well, it gets a slightly negative reward, minus 0.2. 10:53.550 --> 10:57.420 And then last conditions that are related to the reward. 10:57.420 --> 11:00.180 Well, that's if the car is getting too close 11:00.180 --> 11:02.880 to one of the edges, as we spoke of earlier. 11:02.880 --> 11:05.160 Remember when we talked about full sand? 11:05.160 --> 11:07.410 Well, if the car is getting too close 11:07.410 --> 11:11.220 to the left edge of the map, it gets minus one reward. 11:11.220 --> 11:13.620 If it gets too close to the right edge of the map, 11:13.620 --> 11:15.690 it gets reward minus one. 11:15.690 --> 11:18.630 And if it gets too close to the bottom edge of the map, 11:18.630 --> 11:20.340 it gets reward minus one. 11:20.340 --> 11:23.760 And if it gets too close to the upper left of the map, 11:23.760 --> 11:25.230 it gets reward minus one. 11:25.230 --> 11:27.600 So that's a terrible punishment. 11:27.600 --> 11:29.670 And so you will see how it will learn fast 11:29.670 --> 11:31.890 not to rush into some walls. 11:31.890 --> 11:35.250 All right, and then this is to update the goal 11:35.250 --> 11:36.600 when the goal is reached. 11:36.600 --> 11:39.120 So when the car reaches the airport, 11:39.120 --> 11:40.230 which is the first goal, 11:40.230 --> 11:41.940 that is the upper-left corner of the map, 11:41.940 --> 11:43.350 well, the goal changes 11:43.350 --> 11:46.200 to the bottom-right corner of the map, which is downtown. 11:46.200 --> 11:47.550 And that's exactly what we do here. 11:47.550 --> 11:50.070 We update the X-coordinate of the goal 11:50.070 --> 11:51.840 and the Y-coordinate of the goal, 11:51.840 --> 11:55.860 and then we update the distance from the car to the goal. 11:55.860 --> 11:58.320 All right. And then that's less important. 11:58.320 --> 12:01.710 That's just a class that will add the painting tools 12:01.710 --> 12:04.140 for us to be able to paint some roads 12:04.140 --> 12:06.390 or some obstacles on the map. 12:06.390 --> 12:08.070 That's more related to SkyV. 12:08.070 --> 12:09.600 You can have a look if you want. 12:09.600 --> 12:12.690 I'll provide the commented version of this code 12:12.690 --> 12:14.220 and I'll provide some reference 12:14.220 --> 12:17.580 if you want to go deeper on how to do that with SkyV. 12:17.580 --> 12:20.250 But we're getting further from artificial intelligence, 12:20.250 --> 12:22.860 so I'm not gonna go into the details of it. 12:22.860 --> 12:25.320 And that's the same for the last code section here 12:25.320 --> 12:27.060 with the Car app class. 12:27.060 --> 12:31.500 That is just to add the API buttons, clear, save, and load. 12:31.500 --> 12:34.920 So that's what we do here, clear canvas, save. 12:34.920 --> 12:36.150 And that's actually very important. 12:36.150 --> 12:39.240 That's for us to be able to save the AI, 12:39.240 --> 12:42.480 to save the brain so that you can reuse it later 12:42.480 --> 12:44.490 by taking the load function, 12:44.490 --> 12:46.410 which is another tool we add on the map, 12:46.410 --> 12:47.790 to load the brain of the car, 12:47.790 --> 12:49.500 that is, to load the memory of the car, 12:49.500 --> 12:52.230 how to navigate in the map. 12:52.230 --> 12:55.440 And then finally, we have the last of the last code section, 12:55.440 --> 12:56.880 which runs the whole thing, 12:56.880 --> 13:00.900 that is, which runs the map and the AI itself. 13:00.900 --> 13:03.570 And actually, that's what we're gonna do right now. 13:03.570 --> 13:07.020 Let's have a look at everything we make in this code. 13:07.020 --> 13:10.500 So right now, the AI is not implemented 13:10.500 --> 13:13.920 so the car will have a very random movement. 13:13.920 --> 13:16.020 It will actually look like an insect. 13:16.020 --> 13:17.760 But don't worry. We will fix that. 13:17.760 --> 13:21.030 Not only we will train it to move like a real car 13:21.030 --> 13:23.610 and train it to navigate, 13:23.610 --> 13:26.610 following some roads and avoiding some obstacles. 13:26.610 --> 13:27.690 So let's do this. 13:27.690 --> 13:32.520 I'm going to select everything and execute. 13:32.520 --> 13:35.670 And here's the map and here's the car. 13:35.670 --> 13:39.000 All right, so that little thing here that you see 13:39.000 --> 13:41.580 that looks like an insect is our car. 13:41.580 --> 13:46.170 So as I told you, the actions are totally random. 13:46.170 --> 13:50.280 So at each time, the car selects randomly an action 13:50.280 --> 13:53.910 whether to go straight to the left or to the right. 13:53.910 --> 13:57.060 So that's why it is making some nonsense movement 13:57.060 --> 13:59.400 and that's why it's looking like an insect. 13:59.400 --> 14:01.710 So we will fix that, of course. 14:01.710 --> 14:05.640 And of course, since the AI is deactivated, 14:05.640 --> 14:07.290 well, it is not going to the goal, 14:07.290 --> 14:09.240 which is the airport here, 14:09.240 --> 14:13.710 or to downtown at the bottom right of the map. 14:13.710 --> 14:16.020 And we will fix all this by making the AI. 14:16.020 --> 14:21.020 So we will implement the AI into this car or this insect. 14:21.750 --> 14:23.910 So you can see the three balls here, 14:23.910 --> 14:26.580 the yellow one, the red one, and the white one, 14:26.580 --> 14:28.200 that's our three sensors. 14:28.200 --> 14:32.790 So that's what will detect if there is some sand around it. 14:32.790 --> 14:35.700 And speaking of sand, well, let's draw some. 14:35.700 --> 14:40.700 So to do so, I just need to do a click left here. 14:40.950 --> 14:45.480 And drawing some sand by still clicking left. 14:45.480 --> 14:47.970 So right now, I'm adding some sand. 14:47.970 --> 14:50.220 We can add some more. 14:50.220 --> 14:52.740 So each time I'm adding sand, as you can see, 14:52.740 --> 14:55.530 that's putting ones in the sand array. 14:55.530 --> 14:57.090 That's the sand array. 14:57.090 --> 15:00.150 That's the zero, zero coordinates of the origin. 15:00.150 --> 15:02.040 And here there are a lot of ones. 15:02.040 --> 15:04.890 And as you can see, well, that's good to see the car 15:04.890 --> 15:07.590 just went onto the sand and was slowed down. 15:07.590 --> 15:09.480 So as you can see right now, 15:09.480 --> 15:12.480 it is really slowed down because it's going onto the sand 15:12.480 --> 15:14.943 and right now it's trying to escape. 15:16.290 --> 15:19.380 And so what we'll do is we will draw some roads 15:19.380 --> 15:24.180 and we will draw some roads from the airport to downtown 15:24.180 --> 15:26.340 and we will train the car to stay in the road 15:26.340 --> 15:28.440 and to avoid the obstacles. 15:28.440 --> 15:30.840 All right, and as you can see, there is the clear button 15:30.840 --> 15:32.220 to clear the sand. 15:32.220 --> 15:37.220 There is the save button to save the brain of the car. 15:37.290 --> 15:40.230 And actually, there is the score curve that we spoke of. 15:40.230 --> 15:41.430 So that saves the AI, 15:41.430 --> 15:45.030 that saves your model actually, the brain of your car. 15:45.030 --> 15:48.720 And then you can, when you leave your code 15:48.720 --> 15:50.100 or turn off your computer 15:50.100 --> 15:51.870 and you want to go back to it again, 15:51.870 --> 15:54.480 you can use the load button to load your model, 15:54.480 --> 15:56.010 that is, to load the brain, 15:56.010 --> 16:01.010 and that will get the trained AI of your car. 16:01.740 --> 16:06.060 All right, so now I can't wait to start making the AI. 16:06.060 --> 16:07.470 This will be a lot of fun. 16:07.470 --> 16:09.390 We will make our neural network 16:09.390 --> 16:11.340 and we will punish the car 16:11.340 --> 16:14.100 as soon as it doesn't do what we want. 16:14.100 --> 16:16.380 So let's do that from the next tutorial. 16:16.380 --> 16:18.543 And until then, enjoy AI.