1 00:00:00,05 --> 00:00:02,04 - [Instructor] I'm going to teach you how 2 00:00:02,04 --> 00:00:06,00 to understand Edge AI in an autonomous vehicle 3 00:00:06,00 --> 00:00:07,08 through layers of tech. 4 00:00:07,08 --> 00:00:10,08 We'll continue our learning so that you can apply this 5 00:00:10,08 --> 00:00:13,08 to AI anywhere on the web, mobile, 6 00:00:13,08 --> 00:00:16,03 or any Edge device beyond cars. 7 00:00:16,03 --> 00:00:19,06 In this lesson, we will talk about warehouse robots 8 00:00:19,06 --> 00:00:23,00 or autonomous mobility vehicles, AMVs, 9 00:00:23,00 --> 00:00:26,00 and autonomous mobility robots, AMRs, 10 00:00:26,00 --> 00:00:28,09 as they're commonly known in the industry. 11 00:00:28,09 --> 00:00:31,00 So what is Edge AI? 12 00:00:31,00 --> 00:00:33,06 The device can be any IoT device 13 00:00:33,06 --> 00:00:36,04 or an autonomous vehicle, or robotica. 14 00:00:36,04 --> 00:00:39,07 Edge AI means that the device has the processing power 15 00:00:39,07 --> 00:00:42,05 to run the AI in inference on the device, 16 00:00:42,05 --> 00:00:46,00 using the data from the edge device sensors. 17 00:00:46,00 --> 00:00:48,01 This allows for edge computing 18 00:00:48,01 --> 00:00:50,07 or more actions to process the data 19 00:00:50,07 --> 00:00:53,01 by sending to the cloud back and forth 20 00:00:53,01 --> 00:00:55,05 to improve the Edge AI. 21 00:00:55,05 --> 00:00:58,00 The tech stack of an autonomous vehicle 22 00:00:58,00 --> 00:01:02,01 is similar to the layers of Edge AI. 23 00:01:02,01 --> 00:01:06,05 We will soon learn one important distinction. 24 00:01:06,05 --> 00:01:09,00 The autonomous vehicle has many sensors. 25 00:01:09,00 --> 00:01:11,04 These produce data about what it sees 26 00:01:11,04 --> 00:01:13,01 and sensors in the environment. 27 00:01:13,01 --> 00:01:15,06 Sensor fusion is the technology process 28 00:01:15,06 --> 00:01:19,08 of combining the data from multiple sensors to make sense 29 00:01:19,08 --> 00:01:21,06 of the road or the warehouse, 30 00:01:21,06 --> 00:01:24,06 or what the autonomous vehicle is sensing. 31 00:01:24,06 --> 00:01:26,06 This becomes the data layer. 32 00:01:26,06 --> 00:01:28,08 The AI is computer vision 33 00:01:28,08 --> 00:01:30,09 that is the core cognition layers. 34 00:01:30,09 --> 00:01:33,01 This does object detection. 35 00:01:33,01 --> 00:01:36,01 We have learned about this and done a demo, 36 00:01:36,01 --> 00:01:39,06 and we will continue to learn this throughout this course. 37 00:01:39,06 --> 00:01:42,06 The OS layer has more intelligence to use the data 38 00:01:42,06 --> 00:01:46,02 and AI to make the car operate autonomously 39 00:01:46,02 --> 00:01:49,04 with intelligence by utilizing the drive 40 00:01:49,04 --> 00:01:53,00 by wire capability of the vehicle. 41 00:01:53,00 --> 00:01:54,09 Let us look at warehouse robots 42 00:01:54,09 --> 00:01:58,05 or autonomous mobility robots, the AMRs. 43 00:01:58,05 --> 00:02:00,01 We can understand the tech stack 44 00:02:00,01 --> 00:02:02,04 of an autonomous vehicle from this. 45 00:02:02,04 --> 00:02:05,05 Just like Alexa making quick decisions at home, 46 00:02:05,05 --> 00:02:08,08 warehouse robots are making quick decisions on the go, 47 00:02:08,08 --> 00:02:10,07 all without the cloud. 48 00:02:10,07 --> 00:02:12,09 Think about what a warehouse looks like. 49 00:02:12,09 --> 00:02:16,08 When I think of one, I see endless shelves that are sky high 50 00:02:16,08 --> 00:02:20,08 with dozens of boxes or items stacked, aisle numbers 51 00:02:20,08 --> 00:02:22,00 and bright lights. 52 00:02:22,00 --> 00:02:25,06 Every corner we turn that is inventory. 53 00:02:25,06 --> 00:02:28,06 These are the things I notice with my eyes. 54 00:02:28,06 --> 00:02:31,04 For warehouse robots, everything I mentioned before 55 00:02:31,04 --> 00:02:33,06 is picked up by cameras. 56 00:02:33,06 --> 00:02:36,05 These are the cameras that comes with the support 57 00:02:36,05 --> 00:02:38,01 of additional sensors. 58 00:02:38,01 --> 00:02:39,05 The AMR has lidars 59 00:02:39,05 --> 00:02:43,02 and motion sensors to help the AMR navigate 60 00:02:43,02 --> 00:02:45,02 by making sense of its surrounding. 61 00:02:45,02 --> 00:02:46,07 Think of what these sensors are 62 00:02:46,07 --> 00:02:49,06 for your favorite edge device. 63 00:02:49,06 --> 00:02:52,03 These are not your average cameras. 64 00:02:52,03 --> 00:02:55,07 These cameras are connected cameras, 65 00:02:55,07 --> 00:02:59,08 and they work well with connected sensors around them. 66 00:02:59,08 --> 00:03:02,05 These sensors pick up everything in its path 67 00:03:02,05 --> 00:03:05,03 from shells to humans, and every corner 68 00:03:05,03 --> 00:03:08,05 they're collecting and acting on data in real time. 69 00:03:08,05 --> 00:03:11,05 The data from these sensors are able to analyze based 70 00:03:11,05 --> 00:03:15,01 on built-in AI models to recognize shapes, 71 00:03:15,01 --> 00:03:17,08 movement, and distance for example. 72 00:03:17,08 --> 00:03:19,05 This helps the robot understand 73 00:03:19,05 --> 00:03:22,07 its surrounding before moving. 74 00:03:22,07 --> 00:03:25,03 Remember the software layer we discussed? 75 00:03:25,03 --> 00:03:29,00 The software layer for Alexa or Google Home, 76 00:03:29,00 --> 00:03:31,07 what do you think it would be for a robot arm? 77 00:03:31,07 --> 00:03:33,06 Write down your thoughts. 78 00:03:33,06 --> 00:03:35,01 Awesome ideas. 79 00:03:35,01 --> 00:03:36,02 In a robotic arm, 80 00:03:36,02 --> 00:03:39,02 the software layer is acting on the decision of the AI. 81 00:03:39,02 --> 00:03:41,05 They manage all movement or hardware action. 82 00:03:41,05 --> 00:03:44,02 Every wheel turn, arm lift, putting boxes 83 00:03:44,02 --> 00:03:45,07 or packages in a spot. 84 00:03:45,07 --> 00:03:48,01 This is all done by the OS layer. 85 00:03:48,01 --> 00:03:51,01 The OS layer also makes sense of the data 86 00:03:51,01 --> 00:03:54,04 to decide whether the data should stay in the edge 87 00:03:54,04 --> 00:03:55,07 for edge computing, 88 00:03:55,07 --> 00:03:59,09 or should be sent to the cloud to improve the AI. 89 00:03:59,09 --> 00:04:01,05 We don't see the other layers, 90 00:04:01,05 --> 00:04:03,09 but this is the layer which makes sure 91 00:04:03,09 --> 00:04:07,00 that everything is running smoothly and safely. 92 00:04:07,00 --> 00:04:09,02 That is the Edge AI. 93 00:04:09,02 --> 00:04:13,03 The inference layer where the crucial AI decisions are made. 94 00:04:13,03 --> 00:04:17,00 The data that has been cleaned and collected gets analyzed. 95 00:04:17,00 --> 00:04:19,06 Where the model determines if the shelf is 96 00:04:19,06 --> 00:04:22,09 around the corner, or is that a human crossing its path. 97 00:04:22,09 --> 00:04:26,08 This layer is the key to making the AI work. 98 00:04:26,08 --> 00:04:31,02 The three layers of Edge AI, AI, OS, 99 00:04:31,02 --> 00:04:34,08 and data layer works seamlessly to help you make 100 00:04:34,08 --> 00:04:37,00 the most of the device's capability 101 00:04:37,00 --> 00:04:41,00 to solve real business problems using Edge AI.