1 00:00:09,320 --> 00:00:13,910 David Phusion is an amazing tool, which you'll find in pretty much every modern piece of technology 2 00:00:13,910 --> 00:00:19,490 that involves any kind of sensing a measurement or automation, you will find it in your smartphone, 3 00:00:19,490 --> 00:00:20,540 you'll find it in your car. 4 00:00:20,720 --> 00:00:25,970 It has enabled amazing feat of engineering and its usage is only going to dramatically increase with 5 00:00:25,970 --> 00:00:29,680 time with new automation technologies such as yourself driving a car. 6 00:00:30,110 --> 00:00:36,800 Overall, data fusion is the process of integrating multiple data sources together to form a more accurate, 7 00:00:36,950 --> 00:00:41,810 more complete and more robust picture than any one source of information alone could provide. 8 00:00:41,960 --> 00:00:46,910 More specifically, we are going to look at the most common use of data fusion, which is sensor fusion 9 00:00:46,910 --> 00:00:48,230 or multi sensor fusion. 10 00:00:48,290 --> 00:00:53,000 This is the process of how we can take data from sensors and other sources of information and turn it 11 00:00:53,000 --> 00:00:54,170 into useful information. 12 00:00:55,360 --> 00:01:01,600 So now let's look at a practical example, so we have a dynamic system and for this example, let's 13 00:01:01,600 --> 00:01:03,820 take the motion and dynamics of an aircraft. 14 00:01:04,270 --> 00:01:08,980 The state of this system describes the current properties of the aircraft at this specific instance 15 00:01:08,980 --> 00:01:09,600 in time. 16 00:01:09,850 --> 00:01:14,750 So these properties could be things like the velocity, position and attitude of the aircraft. 17 00:01:14,980 --> 00:01:19,540 All of these form the current state of the aircraft or the state of the dynamic system in question. 18 00:01:20,260 --> 00:01:25,840 Modern aircraft systems have order pilots to fly the aircraft, but for the autopilot control to work, 19 00:01:25,840 --> 00:01:30,730 the computer needs to know the current state of the aircraft so that it can calculate how to make the 20 00:01:30,730 --> 00:01:32,320 aircraft respond as required. 21 00:01:35,210 --> 00:01:40,130 To get this information, sensors are attached to the aircraft to measure the physical properties of 22 00:01:40,130 --> 00:01:45,080 the system, these sensors could be things like GPS, speed and altitude sensors. 23 00:01:45,470 --> 00:01:51,440 Each sensor provides measurement data about the state of the aircraft, but in most cases, no one sensor 24 00:01:51,440 --> 00:01:53,600 can sense all the required information. 25 00:01:54,170 --> 00:01:59,690 So to form a complete picture about the estimated state of the aircraft, a sensor fusion process is 26 00:01:59,690 --> 00:02:00,260 required. 27 00:02:03,730 --> 00:02:10,120 Sensor fusion also has more capabilities, no sensor in real life is perfect, any measurement it provides 28 00:02:10,120 --> 00:02:13,300 will be corrupted with errors such as noise and biases. 29 00:02:14,230 --> 00:02:16,960 Fusion allows the effects of these areas to be reduced. 30 00:02:17,290 --> 00:02:23,080 And in many cases, this raw sensor data itself would be useless without television to correct these 31 00:02:23,080 --> 00:02:23,530 errors. 32 00:02:25,040 --> 00:02:31,250 If we take, for example, the speed sensor, if we look at a graph where the Y axis is speed and the 33 00:02:31,250 --> 00:02:37,100 X axis is time, and the white line represents the true speed as it changes with the time, the speed 34 00:02:37,100 --> 00:02:39,290 sensor provides measurements of the speed. 35 00:02:39,290 --> 00:02:43,790 As soon as the crane crosses, you can see that these measurements are noisy. 36 00:02:43,820 --> 00:02:48,710 So if we were to use them directly to estimate the speed, the curve would look like something like 37 00:02:48,710 --> 00:02:50,450 this as seen in the Green Line. 38 00:02:50,870 --> 00:02:54,260 It does not best represent what the true speed actually is doing. 39 00:02:55,490 --> 00:03:01,190 But if we had a bit more information about the sensor, such as how accurate it is or how much error 40 00:03:01,190 --> 00:03:05,780 there might be in each measurement, then we can use this information in the data fusion process. 41 00:03:06,590 --> 00:03:10,730 So the blue bars here represent how much error each measurement might have. 42 00:03:11,060 --> 00:03:14,330 The true value should lie somewhere between the extremes. 43 00:03:15,620 --> 00:03:20,690 So if we were to use this information, along with some basic information about the type of vehicle 44 00:03:20,750 --> 00:03:25,880 the speed sensor is attached to, the data fusion process allows us to get a better estimate of what 45 00:03:25,880 --> 00:03:27,080 the true speed really is. 46 00:03:27,710 --> 00:03:29,870 The red line here is the speed. 47 00:03:30,200 --> 00:03:35,780 You can see here that if you speed is a lot better at approximating the true speed, then using just 48 00:03:35,780 --> 00:03:37,130 the original raw measurements. 49 00:03:38,100 --> 00:03:43,980 This will mean that any automation or control process that uses this information will be more accurate, 50 00:03:43,980 --> 00:03:45,900 more robust and more reliable. 51 00:03:47,240 --> 00:03:52,340 If this was a speed sensor for the cruise control in your car without television, it could be very 52 00:03:52,340 --> 00:03:52,940 dangerous. 53 00:03:52,940 --> 00:03:56,730 It might cause your car to suddenly break or accelerate uncontrollably. 54 00:03:57,680 --> 00:04:02,070 But when data fusion is used, it would allow the sensor to be used very safely. 55 00:04:02,750 --> 00:04:05,000 These days, we just expect things to work. 56 00:04:05,120 --> 00:04:07,100 And without data fusion, they would not.