1 00:00:00,000 --> 00:00:02,200 You know, usually when we talk about getting 2 00:00:02,200 --> 00:00:04,839 a medical diagnosis, there's this expectation 3 00:00:04,839 --> 00:00:08,320 of just absolute, unambiguous precision. Like, 4 00:00:08,599 --> 00:00:11,240 you break your arm, the x -ray shows that jagged 5 00:00:11,240 --> 00:00:13,580 white line on the bone, and the doctor just points 6 00:00:13,580 --> 00:00:15,779 to it. Right. It's right there. Broken or not 7 00:00:15,779 --> 00:00:18,879 broken. Exactly. It's a clean, binary reality. 8 00:00:19,600 --> 00:00:21,920 And, well, I think we find that binary comforting. 9 00:00:22,339 --> 00:00:24,879 We're sort of biologically wired to appreciate 10 00:00:24,879 --> 00:00:27,679 things that are visible and categorizable. With 11 00:00:27,679 --> 00:00:30,510 structurally solid things. Yes, absolutely. But 12 00:00:30,510 --> 00:00:32,509 the moment you step into the world of robotics, 13 00:00:32,770 --> 00:00:36,049 you realize something kind of terrifying. Our 14 00:00:36,049 --> 00:00:37,990 innate human assumption of what the physical 15 00:00:37,990 --> 00:00:41,250 world is, how it functions, how we navigate a 16 00:00:41,250 --> 00:00:44,090 crowded room without bumping into a table, it's 17 00:00:44,090 --> 00:00:47,590 incredibly messy. It is. It's basically a biological 18 00:00:47,590 --> 00:00:50,810 illusion built on millions of years of evolutionary 19 00:00:50,810 --> 00:00:53,530 shortcuts. Right. So to build a machine that 20 00:00:53,530 --> 00:00:55,369 actually understands its environment, you have 21 00:00:55,369 --> 00:00:58,539 to just throw out your biological intuition entirely. 22 00:00:59,320 --> 00:01:01,020 You literally have to build a universe out of 23 00:01:01,020 --> 00:01:04,260 pure mathematics from absolute zero. Because 24 00:01:04,260 --> 00:01:06,519 a robotic system doesn't have an instinctual 25 00:01:06,519 --> 00:01:09,420 sense of up or down. It doesn't inherently know 26 00:01:09,420 --> 00:01:12,640 what a wall is. To engineer a machine that can 27 00:01:12,640 --> 00:01:14,599 interact with the physical world, we first have 28 00:01:14,599 --> 00:01:16,900 to mathematically define what the real world 29 00:01:16,900 --> 00:01:20,439 actually is. Down to the absolute tiniest decimal 30 00:01:20,439 --> 00:01:23,939 point. Exactly. The precise coordinate frame. 31 00:01:24,040 --> 00:01:26,340 And that is exactly what we are undertaking today. 32 00:01:26,500 --> 00:01:28,859 So welcome to the first installment of a special 33 00:01:28,859 --> 00:01:31,459 five -part marginalia series for this deep dive. 34 00:01:32,079 --> 00:01:34,920 Our mission today is to lay the absolute bedrock 35 00:01:34,920 --> 00:01:38,040 for understanding robotics. We are tracing the 36 00:01:38,040 --> 00:01:39,939 evolution of these machines going all the way 37 00:01:39,939 --> 00:01:44,060 from 18th century clockwork mechanisms into the 38 00:01:44,060 --> 00:01:46,359 rigorous spatial mathematics that govern modern 39 00:01:46,359 --> 00:01:49,359 autonomous systems. And to do that, we're focusing 40 00:01:49,359 --> 00:01:52,799 deeply on Peter Corke's Seminal 2023 textbook, 41 00:01:53,219 --> 00:01:56,540 Robotics, Vision, and Control, Fundamental Algorithms 42 00:01:56,540 --> 00:01:58,959 in Python. Such an incredible book. It really 43 00:01:58,959 --> 00:02:02,099 is. Today, for part one of five, we are covering 44 00:02:02,099 --> 00:02:05,120 the introduction and foundations. We are establishing 45 00:02:05,120 --> 00:02:08,180 the fundamental laws of space, time, and motion 46 00:02:08,180 --> 00:02:11,680 that every single robot, from a factory arm to 47 00:02:11,680 --> 00:02:14,819 a Mars rover, has to obey. So for you, the learner 48 00:02:14,819 --> 00:02:17,439 listening to this, whether you're, you know... 49 00:02:17,500 --> 00:02:19,939 Trying to solidify your engineering fundamentals 50 00:02:19,939 --> 00:02:23,560 or you just want to demystify how that self -driving 51 00:02:23,560 --> 00:02:26,159 car next to you is actually processing reality? 52 00:02:26,780 --> 00:02:29,659 This is your starting line. We are going concept 53 00:02:29,659 --> 00:02:32,219 by concept building this mathematical universe 54 00:02:32,219 --> 00:02:34,800 from the ground up. It's a massive intellectual 55 00:02:34,800 --> 00:02:37,360 undertaking, honestly. Because before we even 56 00:02:37,360 --> 00:02:40,419 consider the syntax of programming a robot in 57 00:02:40,419 --> 00:02:43,780 Python, we have to define what a robot is fundamentally. 58 00:02:44,219 --> 00:02:47,080 Right. And to really grasp that definition, it's 59 00:02:47,080 --> 00:02:49,520 incredibly useful to look back at where the human 60 00:02:49,520 --> 00:02:51,840 obsession with creating lifelike machines actually 61 00:02:51,840 --> 00:02:54,300 started. Yeah, and the history here is so much 62 00:02:54,300 --> 00:02:57,000 weirder and honestly more mechanically brilliant 63 00:02:57,000 --> 00:03:00,539 than just circuit boards. Cork takes us back 64 00:03:00,539 --> 00:03:03,259 to the European automata of the 18th century. 65 00:03:03,680 --> 00:03:06,379 And the absolute standout example here is Jacques 66 00:03:06,379 --> 00:03:09,979 de Vaucanson's mechanical duck from 1739. Oh, 67 00:03:09,979 --> 00:03:13,580 the duck. A truly staggering piece of early engineering. 68 00:03:13,840 --> 00:03:17,979 It's wild. It wasn't just a statue. This clockwork 69 00:03:17,979 --> 00:03:20,580 duck could flap its wings with articulated joints, 70 00:03:20,979 --> 00:03:23,419 it could seemingly eat grain from your hand, 71 00:03:24,020 --> 00:03:27,680 and, bizarrely, it could defecate. Yes, the digesting 72 00:03:27,680 --> 00:03:30,520 duck. People in the 18th century looked at this 73 00:03:30,520 --> 00:03:33,180 and genuinely believed they were witnessing something 74 00:03:33,180 --> 00:03:36,080 approaching artificial life. But mechanically, 75 00:03:36,280 --> 00:03:39,110 how was it actually doing all this? So, it relied 76 00:03:39,110 --> 00:03:41,969 on this intensely complex mechanism of rotating 77 00:03:41,969 --> 00:03:45,289 cams. A cam is essentially a shaped wheel, right? 78 00:03:45,550 --> 00:03:48,490 And as it rotates, a follower mechanism traces 79 00:03:48,490 --> 00:03:51,770 its irregular edge. Okay, so it translates rotation 80 00:03:51,770 --> 00:03:55,310 into, like, linear movement? Exactly. In a very 81 00:03:55,310 --> 00:03:57,650 real sense, the physical shape of the cam was 82 00:03:57,650 --> 00:04:00,569 the code. It was a hard -coded program driven 83 00:04:00,569 --> 00:04:03,530 by mechanical clockwork. Bakunsan's duck really 84 00:04:03,530 --> 00:04:05,789 represents the genesis of programmed physical 85 00:04:05,789 --> 00:04:08,360 action. Which introduces one of the most fascinating 86 00:04:08,360 --> 00:04:10,360 technological parallels, I think. Because if 87 00:04:10,360 --> 00:04:13,620 that duck was the equivalent of, say, a single 88 00:04:13,620 --> 00:04:16,160 -use app on your phone, then the jacquard loom 89 00:04:16,160 --> 00:04:19,319 from 1801 was basically the first reprogrammable 90 00:04:19,319 --> 00:04:21,839 physical computer. The jacquard loom is such 91 00:04:21,839 --> 00:04:24,959 a crucial stepping stone. Yeah. Because it used 92 00:04:24,959 --> 00:04:27,459 those stiff, pasteboard -punched cards to weave 93 00:04:27,459 --> 00:04:30,279 incredibly complex patterns into silk. The presence 94 00:04:30,279 --> 00:04:32,279 or absence of a hole in the card determined whether 95 00:04:32,279 --> 00:04:34,620 a specific thread was lifted or lowered. Which 96 00:04:34,620 --> 00:04:37,230 is binary logic. Ones and zeros, but made of 97 00:04:37,230 --> 00:04:40,490 holes and solid cardboard. Exactly. You swap 98 00:04:40,490 --> 00:04:42,389 the stack of cards and the machine completely 99 00:04:42,389 --> 00:04:46,189 changes its physical output. It hit all the foundational 100 00:04:46,189 --> 00:04:49,250 hallmarks of modern robotics, but with wood and 101 00:04:49,250 --> 00:04:52,389 string over a century before the microchip. You 102 00:04:52,389 --> 00:04:55,410 can actually trace a direct technological descent 103 00:04:55,410 --> 00:04:58,850 from those punch cards to Charles Babbage's analytical 104 00:04:58,850 --> 00:05:01,810 engine. And then to Ada Lovelace formalizing 105 00:05:01,810 --> 00:05:04,589 the concept of an algorithm. But despite all 106 00:05:04,589 --> 00:05:07,250 that, the actual word robot didn't even exist 107 00:05:07,250 --> 00:05:10,069 yet. No, it didn't. And the etymology of the 108 00:05:10,069 --> 00:05:13,089 word is deeply revealing about our psychological 109 00:05:13,089 --> 00:05:15,410 relationship with automation. It dates back to 110 00:05:15,410 --> 00:05:18,149 1920. Right, the Czech writer, Paral Chopik. 111 00:05:18,509 --> 00:05:21,509 Yeah, he published a sci -fi play called Rossum's 112 00:05:21,509 --> 00:05:24,850 Universal Robots, or R .U .R. Yeah, and it was 113 00:05:24,850 --> 00:05:27,189 actually his brother, Josef, who suggested the 114 00:05:27,189 --> 00:05:30,569 word roboti. In Czech, the root word robota refers 115 00:05:30,569 --> 00:05:33,610 to servitude. forced labor or the grueling work 116 00:05:33,610 --> 00:05:36,370 of a serf. Which is so telling. And in the play, 117 00:05:36,550 --> 00:05:38,829 these robots aren't clanking metal machines. 118 00:05:39,129 --> 00:05:41,730 They're artificially synthesized organic beings, 119 00:05:42,129 --> 00:05:44,370 essentially androids, made in a factory to do 120 00:05:44,370 --> 00:05:46,410 all the grueling labor humanity didn't want to 121 00:05:46,410 --> 00:05:49,009 do anymore. And, establishing the template for 122 00:05:49,009 --> 00:05:50,949 basically all science fiction for the next century, 123 00:05:51,569 --> 00:05:53,629 these artificial laborers eventually realize 124 00:05:53,629 --> 00:05:56,629 they are being exploited, they violently rebel, 125 00:05:56,990 --> 00:06:00,490 and they eradicate humanity. Classic. But that 126 00:06:00,490 --> 00:06:02,870 underlying cultural anxiety, the fear of the 127 00:06:02,870 --> 00:06:06,170 created subsuming the creator, was fundamentally 128 00:06:06,170 --> 00:06:07,930 baked into the invention of the word itself. 129 00:06:08,490 --> 00:06:11,470 Absolutely. But transitioning from science fiction 130 00:06:11,470 --> 00:06:14,990 to empirical science, Cork highlights the mid 131 00:06:14,990 --> 00:06:17,550 -20th century as the true crucible for modern 132 00:06:17,550 --> 00:06:20,430 robotics. This was the era of the cybernetics 133 00:06:20,430 --> 00:06:23,459 boom. Oh right, pioneered by massive intellects 134 00:06:23,459 --> 00:06:26,279 like Norbert Wiener, Alan Turing, Marvin Minsky. 135 00:06:27,160 --> 00:06:30,100 Cybernetics sounds so retro -futuristic today, 136 00:06:30,240 --> 00:06:32,579 but back then it was radical frontier science. 137 00:06:33,079 --> 00:06:35,779 They were actively blending neurology, control 138 00:06:35,779 --> 00:06:38,240 theory, and information theory. They studied 139 00:06:38,240 --> 00:06:40,860 the human brain not as some mystical organ, but 140 00:06:40,860 --> 00:06:43,660 as an electrical network of neurons using feedback 141 00:06:43,660 --> 00:06:46,800 loops. Asking the dangerous question of whether 142 00:06:46,800 --> 00:06:50,180 we could engineer an electronic analog of a brain. 143 00:06:50,459 --> 00:06:53,160 Which culminated in a huge historical moment 144 00:06:53,160 --> 00:06:56,879 the 1956 Dartmouth conference that legendary 145 00:06:56,879 --> 00:06:59,660 summer research project That's where artificial 146 00:06:59,660 --> 00:07:02,459 intelligence was officially codified as a distinct 147 00:07:02,459 --> 00:07:05,300 field of academic study John McCarthy Marvin 148 00:07:05,300 --> 00:07:08,079 Minsky Claude Shannon They all gathered to formally 149 00:07:08,079 --> 00:07:11,379 define AI and they focused heavily on digital 150 00:07:11,379 --> 00:07:14,660 computers and using logic to simulate reasoning, 151 00:07:14,879 --> 00:07:17,360 which profoundly influenced robotics. It provided 152 00:07:17,360 --> 00:07:19,699 the theoretical brain required to evolve beyond 153 00:07:19,699 --> 00:07:23,300 clockwork cams or simple punched cars. But that 154 00:07:23,300 --> 00:07:25,740 theoretical brain eventually needed a physical 155 00:07:25,740 --> 00:07:28,899 body. Right. Which leads us to the first true 156 00:07:28,899 --> 00:07:32,660 industrial robot. In 1954, George Duvall filed 157 00:07:32,660 --> 00:07:36,120 a patent for universal automation. Duvall's patent 158 00:07:36,120 --> 00:07:38,939 described a track mounted polar coordinate arm 159 00:07:38,939 --> 00:07:41,800 with a gripper governed by a programmable magnetic 160 00:07:41,800 --> 00:07:44,579 drum controller. He then partnered with Joseph 161 00:07:44,579 --> 00:07:47,120 Engelberger, who's universally recognized today 162 00:07:47,120 --> 00:07:50,120 as the father of robotics, to found Unimation. 163 00:07:50,500 --> 00:07:52,459 The world's first robotics company. They took 164 00:07:52,459 --> 00:07:54,899 that theoretical patent and turned it into commercial 165 00:07:54,899 --> 00:07:58,480 iron. By 1961, the Unimate was operating on a 166 00:07:58,480 --> 00:08:00,740 factory floor at a General Motors die casting 167 00:08:00,740 --> 00:08:03,540 plant in New Jersey. Its task was extracting 168 00:08:03,540 --> 00:08:05,899 searing hot metal components from a die casting 169 00:08:05,899 --> 00:08:09,000 machine. An incredibly hazardous, repetitive, 170 00:08:09,199 --> 00:08:12,649 and miserable job for a human. And it used hydraulic 171 00:08:12,649 --> 00:08:16,430 actuators for immense power, right, guided by 172 00:08:16,430 --> 00:08:19,250 a rudimentary electronic memory. It's the absolute 173 00:08:19,250 --> 00:08:21,649 foundation of the modern manufacturing robot. 174 00:08:22,050 --> 00:08:24,370 Yes, and today there are millions of descendants 175 00:08:24,370 --> 00:08:27,050 of the Unimate, spot welding and painting around 176 00:08:27,050 --> 00:08:30,250 the world. But Corey's text gives us a rigorous 177 00:08:30,250 --> 00:08:32,970 taxonomy of robotic forms, because the field 178 00:08:32,970 --> 00:08:35,750 has aggressively evolved way beyond a single 179 00:08:35,750 --> 00:08:38,620 hydraulic arm bolted to a floor. The ecosystem 180 00:08:38,620 --> 00:08:41,580 is wildly diverse now, and a primary distinction 181 00:08:41,580 --> 00:08:44,879 Cork makes is between fixed robots and mobile 182 00:08:44,879 --> 00:08:47,259 robots. Fixed robots are the direct descendants 183 00:08:47,259 --> 00:08:49,860 of the Unimate, the massive manufacturing arms. 184 00:08:50,179 --> 00:08:53,080 Engineered for blinding speed and carrying massive 185 00:08:53,080 --> 00:08:55,899 payloads with sub -millimeter precision, but 186 00:08:55,899 --> 00:08:58,519 they are fundamentally rigid and, more importantly, 187 00:08:58,820 --> 00:09:02,120 they are inherently dangerous. They have zero 188 00:09:02,120 --> 00:09:04,379 situational awareness. Exactly. They operate 189 00:09:04,379 --> 00:09:07,320 inside thick steel safety cages because if a 190 00:09:07,320 --> 00:09:09,139 human accidentally steps into their envelope, 191 00:09:09,480 --> 00:09:11,919 the robot won't stop. It will seamlessly swing 192 00:09:11,919 --> 00:09:14,080 its thousand pound arm right through the human. 193 00:09:14,200 --> 00:09:16,840 Which contrasts sharply with the rapid proliferation 194 00:09:16,840 --> 00:09:20,440 of collaborative robots or cobots. These are 195 00:09:20,440 --> 00:09:22,399 explicitly designed to operate safely in the 196 00:09:22,399 --> 00:09:25,519 exact same workspace as humans, completely without 197 00:09:25,519 --> 00:09:28,139 safety cages. Which begs the question, how do 198 00:09:28,139 --> 00:09:30,379 they achieve that safety mechanically? You can't 199 00:09:30,379 --> 00:09:33,139 just program a robot to be careful. You can't. 200 00:09:33,220 --> 00:09:35,659 They use advanced force torque sensors built 201 00:09:35,659 --> 00:09:38,889 directly into their joints. If the arm encounters 202 00:09:38,889 --> 00:09:41,350 unexpected physical resistance, like brushing 203 00:09:41,350 --> 00:09:44,049 against a person's shoulder, the sensors instantly 204 00:09:44,049 --> 00:09:47,230 detect the spike in force. And the control software 205 00:09:47,230 --> 00:09:50,450 immediately halts or reverses the motion. So 206 00:09:50,450 --> 00:09:53,210 the entire engineering paradigm shifts from rigid 207 00:09:53,210 --> 00:09:57,330 isolation to safe, compliant collaboration. Exactly. 208 00:09:57,870 --> 00:10:00,570 Then we venture into mobile robots, which are 209 00:10:00,570 --> 00:10:02,669 generally categorized by the specific environment 210 00:10:02,669 --> 00:10:05,500 they navigate. On the ground, you have robots 211 00:10:05,500 --> 00:10:07,779 using wheels, like Amazon's autonomous guided 212 00:10:07,779 --> 00:10:11,879 vehicles, or legs, like Boston Dynamics' quadruped 213 00:10:11,879 --> 00:10:14,860 spot. And moving upward, we have uncrewed aerial 214 00:10:14,860 --> 00:10:18,120 vehicles, or UAVs, everything from small quadcopters 215 00:10:18,120 --> 00:10:20,419 to massive fixed -wing systems like the Global 216 00:10:20,419 --> 00:10:23,039 Hawk. Beneath the surface, there are autonomous 217 00:10:23,039 --> 00:10:25,960 underwater vehicles, AUVs, mapping the ocean 218 00:10:25,960 --> 00:10:29,210 floor. And in the most hostile environment imaginable, 219 00:10:29,690 --> 00:10:31,629 space robotics, like the Perseverance Rover, 220 00:10:31,850 --> 00:10:34,730 doing autonomous science on Mars right now. It 221 00:10:34,730 --> 00:10:37,970 is a stunning menagerie of machines. But Cork 222 00:10:37,970 --> 00:10:40,509 emphasizes another crucial category that bridges 223 00:10:40,509 --> 00:10:43,929 human and machine intelligence. Telerobots. Right. 224 00:10:43,990 --> 00:10:47,190 Robotic systems operated from a physical distance 225 00:10:47,190 --> 00:10:50,570 by a human pilot. And the genesis of telerobotics 226 00:10:50,570 --> 00:10:52,610 is actually rooted in the Manhattan Project during 227 00:10:52,610 --> 00:10:55,340 World War II. Scientists needed to manipulate 228 00:10:55,340 --> 00:10:58,120 highly radioactive isotopes, but direct human 229 00:10:58,120 --> 00:11:00,980 contact was obviously lethal. So they developed 230 00:11:00,980 --> 00:11:04,340 mechanical master -slave manipulators. The human 231 00:11:04,340 --> 00:11:07,139 scientists stood safely behind feet of lead glass, 232 00:11:07,159 --> 00:11:10,740 moving a complex mechanical master arm. Cables 233 00:11:10,740 --> 00:11:12,779 and pulleys transmitted those movements through 234 00:11:12,779 --> 00:11:15,360 the wall to a slave arm inside the hot cell, 235 00:11:15,820 --> 00:11:18,279 replicating the motions perfectly. It effectively 236 00:11:18,279 --> 00:11:21,320 projected human physical presence across a hazardous 237 00:11:21,320 --> 00:11:23,899 barrier. And we see this today with remotely 238 00:11:23,899 --> 00:11:26,500 operated vehicles or ROVs exploring the ocean 239 00:11:26,500 --> 00:11:29,460 floor that used to find the Titanic. And it's 240 00:11:29,460 --> 00:11:32,220 revolutionizing medicine, too. Companies like 241 00:11:32,220 --> 00:11:34,779 Intuitive Surgical make immensely complex surgical 242 00:11:34,779 --> 00:11:37,399 telerobots. The surgeon doesn't stand over the 243 00:11:37,399 --> 00:11:39,299 patient. They sit at a highly advanced console 244 00:11:39,299 --> 00:11:42,759 across the room looking into a 3D display. Moving 245 00:11:42,759 --> 00:11:46,059 ergonomic controllers and the robotic arms. Do 246 00:11:46,059 --> 00:11:48,279 the surgery inside the patient through microscopic 247 00:11:48,279 --> 00:11:50,860 incisions, completely filtering out natural human 248 00:11:50,860 --> 00:11:53,259 hand tremors. It's incredible. But this intricate 249 00:11:53,259 --> 00:11:56,139 interaction between a human operator and a robotic 250 00:11:56,139 --> 00:11:59,179 system forces us to confront a vital engineering 251 00:11:59,179 --> 00:12:01,879 distinction about the control loop. When a human 252 00:12:01,879 --> 00:12:04,480 and a machine collaborate, who actually possesses 253 00:12:04,480 --> 00:12:07,000 authority over the machine at any given millisecond? 254 00:12:07,659 --> 00:12:10,299 Quirk provides a very strict delineation here 255 00:12:10,299 --> 00:12:13,580 between traded control and shared control. This 256 00:12:13,580 --> 00:12:15,379 is a profound concept to wrap your head around, 257 00:12:15,600 --> 00:12:17,720 especially as automation seeps into everyday 258 00:12:17,720 --> 00:12:20,860 life. Traded control is sequential. The authority 259 00:12:20,860 --> 00:12:23,299 is passed back and forth. Think of a relay race. 260 00:12:23,940 --> 00:12:26,759 A human pilot flies an aircraft, makes a conscious 261 00:12:26,759 --> 00:12:29,700 decision, flips a switch, and hands the baton 262 00:12:29,700 --> 00:12:32,779 to the autopilot. The machine is now entirely 263 00:12:32,779 --> 00:12:35,460 in control. The human is just a passive monitor 264 00:12:35,460 --> 00:12:37,779 until they take the baton back. It's an exclusive 265 00:12:37,779 --> 00:12:40,539 exchange. Only one entity is driving at a time. 266 00:12:41,299 --> 00:12:44,779 Shared control, however, is a simultaneous blended 267 00:12:44,779 --> 00:12:47,639 action. The human and the machine exert authority 268 00:12:47,639 --> 00:12:50,379 over the physical system at the exact same moment. 269 00:12:50,919 --> 00:12:53,360 Like lane keeping assist in a modern car. You 270 00:12:53,360 --> 00:12:55,620 are dictating the speed with the pedal. You are 271 00:12:55,620 --> 00:12:58,159 physically holding the steering wheel. But simultaneously, 272 00:12:58,340 --> 00:13:00,559 the car's computer vision is actively tracking 273 00:13:00,559 --> 00:13:03,639 the white lines. And if you drift slightly, the 274 00:13:03,639 --> 00:13:05,960 computer exerts a measured torque on the steering 275 00:13:05,960 --> 00:13:09,559 column, nudging you back to the center. You feel... 276 00:13:09,580 --> 00:13:11,860 the computer's physical input through your hands. 277 00:13:12,340 --> 00:13:15,220 You are operating it concurrently. That blended 278 00:13:15,220 --> 00:13:17,860 interaction is the hallmark of shared control. 279 00:13:18,480 --> 00:13:21,679 But regardless of the form factory, ARM, Deepsea 280 00:13:21,679 --> 00:13:25,700 ROV, or a smart car, we have to ask, what unifies 281 00:13:25,700 --> 00:13:28,759 them? How do we define what a robot actually 282 00:13:28,759 --> 00:13:31,659 is? Cork provides a foundational definition that 283 00:13:31,659 --> 00:13:34,159 anchors the entire textbook. He defines a robot 284 00:13:34,159 --> 00:13:37,820 as a goal -oriented machine that can sense, plan, 285 00:13:37,820 --> 00:13:42,740 and act. Sense. Plan. Act. The fundamental control 286 00:13:42,740 --> 00:13:45,840 loop. It seems simple, but unpacking it reveals 287 00:13:45,840 --> 00:13:48,659 immense complexity. We have to dismantle the 288 00:13:48,659 --> 00:13:51,779 sense portion first. Because a robot is utterly 289 00:13:51,779 --> 00:13:54,000 useless if it's ignorant of its own physical 290 00:13:54,000 --> 00:13:56,440 state and its surroundings. Cork categorizes 291 00:13:56,440 --> 00:13:58,700 robotic perception into two distinct sensory 292 00:13:58,700 --> 00:14:01,840 domains. The first is proprioceptive sensing. 293 00:14:02,500 --> 00:14:05,309 Like human proprioception. your innate awareness 294 00:14:05,309 --> 00:14:08,009 of your body in space. Even with your eyes closed, 295 00:14:08,049 --> 00:14:10,330 you know the exact angle of your elbow. For a 296 00:14:10,330 --> 00:14:12,690 robot, these sensors are strictly internal, right? 297 00:14:13,090 --> 00:14:15,350 Yes, measuring the robot's own physical state. 298 00:14:16,539 --> 00:14:19,120 Mechanically, this involves rotary encoders mounted 299 00:14:19,120 --> 00:14:22,340 on motor shafts to measure joint angles down 300 00:14:22,340 --> 00:14:25,299 to fractions of a degree, or current sensors 301 00:14:25,299 --> 00:14:27,840 monitoring electrical draw to determine how much 302 00:14:27,840 --> 00:14:30,539 physical torque it's exerting. The robot is constantly 303 00:14:30,539 --> 00:14:33,919 looking inward. Right. Then the second domain, 304 00:14:34,139 --> 00:14:37,480 is exteroceptive sensing, which is the robot 305 00:14:37,480 --> 00:14:39,840 looking outward trying to comprehend the external 306 00:14:39,840 --> 00:14:42,340 universe. And these range from extreme simplicity 307 00:14:42,340 --> 00:14:45,679 to staggering complexity. A mechanical microswitch 308 00:14:45,679 --> 00:14:48,419 on a Roomba bumper is an exteroceptive sensor. 309 00:14:49,220 --> 00:14:51,580 It registers a binary collision with a wall. 310 00:14:51,750 --> 00:14:55,049 A GPS receiver pinging satellites is exteroceptive. 311 00:14:55,470 --> 00:14:57,950 A LIDAR system spinning a laser beam to generate 312 00:14:57,950 --> 00:15:00,870 a point cloud is exteroceptive. Yeah, but Cork 313 00:15:00,870 --> 00:15:03,210 posits throughout the text that the ultimate 314 00:15:03,210 --> 00:15:07,100 exteroceptive sensor, the most data -rich method 315 00:15:07,100 --> 00:15:10,700 to parse the world, is vision. Okay, let's unpack 316 00:15:10,700 --> 00:15:14,580 this because robotic vision is a problem of staggering 317 00:15:14,580 --> 00:15:17,299 mathematical complexity. We underestimate it 318 00:15:17,299 --> 00:15:20,179 because biological organisms do it so seamlessly. 319 00:15:20,419 --> 00:15:24,269 It feels effortless to us. But Cork emphasizes 320 00:15:24,269 --> 00:15:27,129 the evolutionary effort required to develop sight. 321 00:15:27,769 --> 00:15:30,889 Nature didn't just invent the eye once. Evolutionary 322 00:15:30,889 --> 00:15:33,730 biologists believe complex eyes evolved independently 323 00:15:33,730 --> 00:15:36,610 at least seven different times across the history 324 00:15:36,610 --> 00:15:40,090 of life on Earth. Resulting in like 10 entirely 325 00:15:40,090 --> 00:15:42,870 distinct eptical architectures across species. 326 00:15:43,049 --> 00:15:45,470 It's the most highly optimized biological feature 327 00:15:45,470 --> 00:15:47,809 in existence. Consider a honeybee. Its brain 328 00:15:47,809 --> 00:15:51,509 has roughly one million neurons. microscopic 329 00:15:51,509 --> 00:15:54,649 computing engine. Yet, relying almost entirely 330 00:15:54,649 --> 00:15:57,389 on vision, it navigates complex environments, 331 00:15:58,049 --> 00:16:00,809 finds a flower miles away, and flawlessly returns 332 00:16:00,809 --> 00:16:03,090 to a tiny opening in its hive. Contrast that 333 00:16:03,090 --> 00:16:06,929 with humans. Our visual processing is so computationally 334 00:16:06,929 --> 00:16:08,929 demanding that it consumes roughly one -third 335 00:16:08,929 --> 00:16:11,889 of our entire brain's processing capacity. We 336 00:16:11,889 --> 00:16:14,629 dedicate a staggering volume of biological hardware 337 00:16:14,629 --> 00:16:17,730 just to interpreting photons bouncing off geometry, 338 00:16:18,250 --> 00:16:20,269 which perfectly illustrates the challenge of 339 00:16:20,269 --> 00:16:23,269 computer vision. The sensory hardware is trivial. 340 00:16:23,409 --> 00:16:25,850 You can buy a CMOS camera chip for a few dollars. 341 00:16:26,250 --> 00:16:29,230 But programming a microprocessor to semantically 342 00:16:29,230 --> 00:16:31,330 understand the pixel data from that camera is 343 00:16:31,330 --> 00:16:34,190 a monumental hurdle. Why is it so mathematically 344 00:16:34,190 --> 00:16:36,429 difficult? Because of the physics of projection. 345 00:16:37,070 --> 00:16:40,529 A standard camera takes a rich 3D physical universe 346 00:16:41,120 --> 00:16:43,919 and violently compresses it through a lens, projecting 347 00:16:43,919 --> 00:16:47,860 it onto a flat 2D grid of pixels. In that transformation, 348 00:16:48,139 --> 00:16:51,960 the entire z -axis, the depth, is instantly and 349 00:16:51,960 --> 00:16:54,659 completely obliterated. So the camera knows the 350 00:16:54,659 --> 00:16:56,919 x and y position of a coffee cup in the image, 351 00:16:56,980 --> 00:16:59,460 but it has no idea if the cup is 2 feet away 352 00:16:59,460 --> 00:17:02,519 or 20 feet away. Precisely. The depth ray is 353 00:17:02,519 --> 00:17:05,500 lost. Okay, acting on behalf of the pragmatic 354 00:17:05,500 --> 00:17:08,839 engineers listening. If computer vision is this 355 00:17:08,839 --> 00:17:10,799 computationally expensive and a camera destroys 356 00:17:10,799 --> 00:17:13,039 all depth of information anyway, why bother? 357 00:17:13,200 --> 00:17:16,859 Why not just preload a flawless millimeter -accurate 358 00:17:16,859 --> 00:17:19,559 3D laser map of the environment, use GPS, and 359 00:17:19,559 --> 00:17:21,819 let the robot navigate blindfolded based on the 360 00:17:21,819 --> 00:17:24,339 map? It's a logical proposition. And it would 361 00:17:24,339 --> 00:17:26,579 function perfectly if the universe were completely 362 00:17:26,579 --> 00:17:29,440 static. If the environment never changed, a blind 363 00:17:29,440 --> 00:17:32,099 robot on a perfect map would be fine. But the 364 00:17:32,099 --> 00:17:35,079 physical world is chaotic, stochastic, and constantly 365 00:17:35,079 --> 00:17:38,210 evolving. Right. A preloaded city map can't tell 366 00:17:38,210 --> 00:17:40,349 a self -driving car that a child just chased 367 00:17:40,349 --> 00:17:42,970 a ball under the street. A hospital map can't 368 00:17:42,970 --> 00:17:45,170 alert a delivery robot that someone left a linen 369 00:17:45,170 --> 00:17:47,869 cart blocking a hallway. The map is a historical 370 00:17:47,869 --> 00:17:50,410 record of the past. The robot must survive in 371 00:17:50,410 --> 00:17:52,730 the present reality. It has to look outward. 372 00:17:53,410 --> 00:17:55,490 And to solve that flattened 2D image crisis, 373 00:17:55,950 --> 00:17:58,829 roboticists frequently use stereo vision. Copying 374 00:17:58,829 --> 00:18:01,930 biology again. Like humans using two eyes spaced 375 00:18:01,930 --> 00:18:04,440 apart to perceive depth. Engineers mount two 376 00:18:04,440 --> 00:18:06,579 cameras with a known physical baseline between 377 00:18:06,579 --> 00:18:10,259 them. The control software captures two 2D images 378 00:18:10,259 --> 00:18:13,059 simultaneously. Because the cameras are offset, 379 00:18:13,619 --> 00:18:15,759 an object will appear at slightly different pixel 380 00:18:15,759 --> 00:18:19,000 coordinates in the left image versus the right. 381 00:18:19,359 --> 00:18:22,059 The software identifies identical features and 382 00:18:22,059 --> 00:18:24,940 calculates the horizontal pixel shift or disparity. 383 00:18:25,289 --> 00:18:27,970 And using basic trigonometry, if the computer 384 00:18:27,970 --> 00:18:30,430 knows the exact distance between the lenses and 385 00:18:30,430 --> 00:18:33,470 the disparity, it can triangulate the exact distance 386 00:18:33,470 --> 00:18:36,150 to the object. Effectively clawing back that 387 00:18:36,150 --> 00:18:39,269 lost depth structure. This is exactly why the 388 00:18:39,269 --> 00:18:41,829 camera mast on the Mars Perseverance rover looks 389 00:18:41,829 --> 00:18:44,809 like a robotic head with two eyes. It's constantly 390 00:18:44,809 --> 00:18:47,990 using stereo triangulation to calculate the depth 391 00:18:47,990 --> 00:18:50,849 of craters and boulders. Which introduces a critical 392 00:18:50,849 --> 00:18:53,519 transition in Cork's framework. Once you engineer 393 00:18:53,519 --> 00:18:56,140 a machine that can sense, plan a trajectory, 394 00:18:56,380 --> 00:18:58,839 and act on the world, you've unleashed an autonomous 395 00:18:58,839 --> 00:19:01,599 agent into human society. And the moment that 396 00:19:01,599 --> 00:19:04,759 happens, robotics ceases to be just an engineering 397 00:19:04,759 --> 00:19:08,039 discipline. It becomes a profound societal and 398 00:19:08,039 --> 00:19:10,980 ethical dilemma. Cork dedicates a substantial 399 00:19:10,980 --> 00:19:13,599 portion of Chapter One to the concept of the 400 00:19:13,599 --> 00:19:16,819 ethical operator. He addresses the pervasive 401 00:19:16,819 --> 00:19:19,380 societal anxieties, emphasizing that engineers 402 00:19:19,380 --> 00:19:21,500 have to evaluate these issues with objective 403 00:19:21,500 --> 00:19:24,380 scrutiny, not emotional reactions. Let's look 404 00:19:24,380 --> 00:19:27,059 at the most prominent anxiety, the debate over 405 00:19:27,059 --> 00:19:30,279 automation taking human jobs. Cork meticulously 406 00:19:30,279 --> 00:19:32,779 outlines the dichotomy here. On one side, it 407 00:19:32,779 --> 00:19:35,819 is an undeniable empirical reality that automation 408 00:19:35,819 --> 00:19:38,940 threatens routine manual labor. As robots become 409 00:19:38,940 --> 00:19:41,690 cheaper and more capable, Repetitive physical 410 00:19:41,690 --> 00:19:44,390 tasks will inevitably transition to systems that 411 00:19:44,390 --> 00:19:47,849 work continuously without fatigue. But on the 412 00:19:47,849 --> 00:19:50,109 other hand, Cork presents the counter perspective 413 00:19:50,109 --> 00:19:52,910 regarding critical labor shortages. There are 414 00:19:52,910 --> 00:19:56,309 sectors involving brutal degrading labor facing 415 00:19:56,309 --> 00:19:59,309 systemic shortages of human workers. Agricultural 416 00:19:59,309 --> 00:20:02,089 harvesting is a prime example. The manual labor 417 00:20:02,089 --> 00:20:04,670 of picking fruit is physically punishing. If 418 00:20:04,670 --> 00:20:08,019 they can't recruit human labor, crops rot. food 419 00:20:08,019 --> 00:20:10,319 costs skyrocket or production moves to other 420 00:20:10,319 --> 00:20:13,000 countries, drastically increasing the environmental 421 00:20:13,000 --> 00:20:16,019 footprint of shipping. In that context, a fruit 422 00:20:16,019 --> 00:20:19,180 picking robot isn't eliminating a job, it's securing 423 00:20:19,180 --> 00:20:21,880 the food supply chain. Furthermore, Quark highlights 424 00:20:21,880 --> 00:20:24,819 global manufacturing realities. In high wage 425 00:20:24,819 --> 00:20:27,220 economies, the U .S., Western Europe, Japan, 426 00:20:27,980 --> 00:20:30,660 advanced robotics has paradoxically kept manufacturing 427 00:20:30,660 --> 00:20:33,579 sectors economically viable. Because the alternative 428 00:20:33,579 --> 00:20:35,660 wasn't keeping the human assembly line workers. 429 00:20:35,720 --> 00:20:38,440 The alternative was the entire factory closing 430 00:20:38,440 --> 00:20:40,900 and relocating to a low -wage country. Exactly. 431 00:20:41,200 --> 00:20:42,900 Automation reduced the sheer volume of human 432 00:20:42,900 --> 00:20:46,279 line workers required. But the massive increase 433 00:20:46,279 --> 00:20:49,359 in efficiency reduced the unit cost, allowing 434 00:20:49,359 --> 00:20:52,079 the factory to remain competitive globally. It 435 00:20:52,079 --> 00:20:54,420 preserved the industry, maintaining the engineering 436 00:20:54,420 --> 00:20:57,099 jobs, maintenance roles, and the local tax base 437 00:20:57,099 --> 00:20:58,720 that would have otherwise vanished entirely. 438 00:20:59,430 --> 00:21:02,359 It's a highly nuanced economic balance. And the 439 00:21:02,359 --> 00:21:04,599 ethical complexities deepen when we look at physical 440 00:21:04,599 --> 00:21:07,359 safety and the psychology of blame. Cork points 441 00:21:07,359 --> 00:21:10,319 out a massive hypocrisy in how society assesses 442 00:21:10,319 --> 00:21:13,859 risk. The statistics are sobering. Human drivers 443 00:21:13,859 --> 00:21:15,940 cause accidents resulting in over one million 444 00:21:15,940 --> 00:21:19,180 fatalities globally every single year. One million 445 00:21:19,180 --> 00:21:22,700 people. Yeah. Yet we've normalized it as an inherent 446 00:21:22,700 --> 00:21:25,880 risk of modern life. But with autonomous vehicles, 447 00:21:26,220 --> 00:21:28,559 human psychology reacts totally differently. 448 00:21:28,990 --> 00:21:32,009 Society exhibits visceral discomfort with self 449 00:21:32,009 --> 00:21:34,410 -driving cars, even when data shows they can 450 00:21:34,410 --> 00:21:37,450 drastically reduce collisions. We have an incredibly 451 00:21:37,450 --> 00:21:40,609 high tolerance for human error, but demand absolute 452 00:21:40,609 --> 00:21:43,049 perfection from machines. When a human makes 453 00:21:43,049 --> 00:21:45,970 a mistake, we blame the individual. They were 454 00:21:45,970 --> 00:21:49,109 distracted. When an autonomous vehicle crashes, 455 00:21:49,690 --> 00:21:51,829 public discourse immediately questions the viability 456 00:21:51,829 --> 00:21:54,230 of the entire technology. And the philosophy 457 00:21:54,230 --> 00:21:56,670 gets even more intimate with robotic caregiving. 458 00:21:57,200 --> 00:21:59,640 We're seeing robots designed for elder care or 459 00:21:59,640 --> 00:22:01,400 assisting people with physical disabilities. 460 00:22:01,779 --> 00:22:03,660 And from a pure engineering perspective, it's 461 00:22:03,660 --> 00:22:06,220 solvable. A robot can safely fetch medication 462 00:22:06,220 --> 00:22:09,160 and monitor vital signs. But the profound ethical 463 00:22:09,160 --> 00:22:12,680 question Cork raises is the human cost. Does 464 00:22:12,680 --> 00:22:16,039 replacing a human nurse with an autonomous machine 465 00:22:16,039 --> 00:22:18,339 fundamentally degrade the patient's quality of 466 00:22:18,339 --> 00:22:21,220 life? Right. A robot can dispense a pill, but 467 00:22:21,220 --> 00:22:23,619 it can't provide genuine empathy or authentic 468 00:22:23,619 --> 00:22:26,299 companionship. Are we prioritizing mechanical 469 00:22:26,299 --> 00:22:29,859 efficiency over human dignity? These aren't abstract 470 00:22:29,859 --> 00:22:33,660 experiments. They're immediate realities. Cork 471 00:22:33,660 --> 00:22:35,660 argues emphatically that engineers can't just 472 00:22:35,660 --> 00:22:38,180 hide in a lab, building systems to see if the 473 00:22:38,180 --> 00:22:40,799 math works. They must be active participants 474 00:22:40,799 --> 00:22:43,940 in these ethical debates. But to actually engineer 475 00:22:43,940 --> 00:22:46,720 systems robust enough to navigate these challenges, 476 00:22:47,180 --> 00:22:49,559 roboticists need powerful mathematical tools. 477 00:22:50,079 --> 00:22:52,400 Transitioning from the philosophical to the highly 478 00:22:52,400 --> 00:22:54,900 practical, Cork introduces the core methodology 479 00:22:54,900 --> 00:22:58,339 of his textbook. He is a massive proponent of 480 00:22:58,339 --> 00:23:02,079 utilizing Python. Yes, specifically several specialized 481 00:23:02,079 --> 00:23:05,240 Python toolboxes he developed over decades. The 482 00:23:05,240 --> 00:23:08,000 spatial maths toolbox, the robotics toolbox, 483 00:23:08,599 --> 00:23:12,039 machine vision and block diagram toolboxes. Why 484 00:23:12,039 --> 00:23:14,299 Python, though? Why not a lower level language 485 00:23:14,299 --> 00:23:17,980 like C++ that might execute faster? The pedagogical 486 00:23:17,980 --> 00:23:20,559 philosophy of the book is rapid prototyping and 487 00:23:20,559 --> 00:23:23,460 instantaneous gratification. Cork knows spatial 488 00:23:23,460 --> 00:23:25,859 mathematics is intensely abstract and hard to 489 00:23:25,859 --> 00:23:28,180 conceptualize on a whiteboard. He wants you to 490 00:23:28,180 --> 00:23:30,500 type a few lines of code, execute the script, 491 00:23:30,759 --> 00:23:33,940 and immediately see a 3D visualization of a robotic 492 00:23:33,940 --> 00:23:37,079 arm dynamically moving on your screen. He believes 493 00:23:37,079 --> 00:23:39,200 theoretical knowledge only truly solidifies when 494 00:23:39,200 --> 00:23:42,359 you actively apply it and see the simulated physical 495 00:23:42,359 --> 00:23:45,119 results. And that framework concludes our exploration 496 00:23:45,119 --> 00:23:47,799 of the introductory concepts. We have the definition, 497 00:23:48,160 --> 00:23:50,880 the sensory loops, and the ethical weight. Now 498 00:23:50,880 --> 00:23:53,339 we transition into the core physics. We have 499 00:23:53,339 --> 00:23:55,160 to construct the mathematical universe these 500 00:23:55,160 --> 00:23:57,819 machines operate in. We're leaving the historical 501 00:23:57,819 --> 00:24:00,140 context and plunging directly into the rigorous 502 00:24:00,140 --> 00:24:03,220 spatial mathematics of Chapter 2. The foundational 503 00:24:03,220 --> 00:24:05,700 problem we have to solve is this. If we're going 504 00:24:05,700 --> 00:24:07,720 to program a robot to interact with a physical 505 00:24:07,720 --> 00:24:10,480 object, we must first have a mathematical language 506 00:24:10,480 --> 00:24:12,539 to tell the robot where that object actually 507 00:24:12,539 --> 00:24:15,990 is. We must establish the geometry of where. 508 00:24:16,490 --> 00:24:18,349 And this begins with the single most critical 509 00:24:18,349 --> 00:24:21,509 concept in all of robotics, the coordinate frame. 510 00:24:22,369 --> 00:24:24,589 I want to spend significant time on this because 511 00:24:24,589 --> 00:24:27,809 everything else builds on it. How do we mathematically 512 00:24:27,809 --> 00:24:30,750 define a coordinate frame? Imagine you take a 513 00:24:30,750 --> 00:24:33,089 physical object, let's say a robotic gripper, 514 00:24:34,029 --> 00:24:37,009 mentally attach a rigid set of three intersecting 515 00:24:37,009 --> 00:24:39,869 arrows to the center of that gripper. These arrows 516 00:24:39,869 --> 00:24:43,839 represent the X, Y, and Z axes. The exact point 517 00:24:43,839 --> 00:24:47,000 where their tails meet is the origin. And these 518 00:24:47,000 --> 00:24:50,200 axes aren't pointing randomly. They must be orthogonal, 519 00:24:50,579 --> 00:24:53,559 perfect 90 -degree angles to each other. Furthermore, 520 00:24:54,000 --> 00:24:56,359 in robotics, we strictly adhere to a convention 521 00:24:56,359 --> 00:24:59,140 called the right -handed coordinate system. To 522 00:24:59,140 --> 00:25:01,559 visualize this, hold up your right hand. Point 523 00:25:01,559 --> 00:25:03,940 your index finger straight forward. That's the 524 00:25:03,940 --> 00:25:06,859 x -axis. Point your middle finger to the left, 525 00:25:07,299 --> 00:25:09,380 perpendicular to your index finger. That's the 526 00:25:09,380 --> 00:25:13,230 y -axis. Now... point your thumb straight up. 527 00:25:13,410 --> 00:25:16,210 Perpendicular to both fingers, that is the z 528 00:25:16,210 --> 00:25:19,250 -axis. This right -hand rule ensures that the 529 00:25:19,250 --> 00:25:21,630 math of cross products and rotations remains 530 00:25:21,630 --> 00:25:25,529 consistent globally. Once you've glued this invisible 531 00:25:25,529 --> 00:25:28,190 right -handed frame to the gripper, you can define 532 00:25:28,190 --> 00:25:30,829 its state relative to another frame, like one 533 00:25:30,829 --> 00:25:33,710 attached to a table. We refer to this relative 534 00:25:33,710 --> 00:25:37,769 state as its pose. And pose carries a very strict 535 00:25:37,769 --> 00:25:40,630 mathematical definition here. Pose is not just 536 00:25:40,630 --> 00:25:44,720 location. Pose equals position plus orientation. 537 00:25:45,500 --> 00:25:48,619 Position is pure translation. How far the origin 538 00:25:48,619 --> 00:25:51,480 has moved along x, y, and z relative to the table. 539 00:25:51,980 --> 00:25:55,019 Just three numbers representing distance. Orientation, 540 00:25:55,099 --> 00:25:58,420 however, is the attitude. It describes how much 541 00:25:58,420 --> 00:26:00,599 the gripper's frame has rotated around those 542 00:26:00,599 --> 00:26:03,519 x, y, and z axes. Because you could have the 543 00:26:03,519 --> 00:26:06,299 exact correct XYZ coordinates to grab a coffee 544 00:26:06,299 --> 00:26:09,740 cup, but if the orientation is rotated 180 degrees, 545 00:26:09,839 --> 00:26:12,339 meaning it's upside down, the robot will smash 546 00:26:12,339 --> 00:26:15,119 the cup instead of picking it up. Exactly. Pose 547 00:26:15,119 --> 00:26:18,240 inherently requires both translation and rotation. 548 00:26:19,240 --> 00:26:22,240 Once you grasp that, Cork introduces a phenomenal 549 00:26:22,240 --> 00:26:25,240 conceptual tool for organizing this data, the 550 00:26:25,240 --> 00:26:28,640 Pose Graph. He asks us to visualize the physical 551 00:26:28,640 --> 00:26:31,369 environment as a network diagram. So consider 552 00:26:31,369 --> 00:26:33,910 a manufacturing work cell. You have the robot 553 00:26:33,910 --> 00:26:36,589 base bolted to the floor, the gripper at the 554 00:26:36,589 --> 00:26:39,049 end of the arm, a digital camera on a tripod, 555 00:26:39,269 --> 00:26:42,109 and a metal widget on a conveyor belt. In a pose 556 00:26:42,109 --> 00:26:44,789 graph, every single one of those entities is 557 00:26:44,789 --> 00:26:47,509 a vertex, a distinct node in the mathematical 558 00:26:47,509 --> 00:26:50,289 network. And the edges, the lines connecting 559 00:26:50,289 --> 00:26:52,829 those nodes, represent the relative poses between 560 00:26:52,829 --> 00:26:55,529 them, the mathematical transformations. Yes. 561 00:26:56,109 --> 00:26:58,509 There's an edge from the camera to the widget. 562 00:26:58,680 --> 00:27:00,819 representing the distance and angle between them. 563 00:27:01,220 --> 00:27:03,720 There's an edge from the robot base to the gripper. 564 00:27:04,359 --> 00:27:06,480 The absolute genius of the pose graph is that 565 00:27:06,480 --> 00:27:08,859 it allows you to compose relative motions through 566 00:27:08,859 --> 00:27:11,740 matrix multiplication. If your software knows 567 00:27:11,740 --> 00:27:14,240 the pose of the camera relative to the base, 568 00:27:14,460 --> 00:27:16,619 and the camera calculates the pose of the widget 569 00:27:16,619 --> 00:27:19,059 relative to its lens, you can mathematically 570 00:27:19,059 --> 00:27:21,619 chain those transformations together. You walk 571 00:27:21,619 --> 00:27:24,420 along the edges of the graph. multiply the transformation 572 00:27:24,420 --> 00:27:27,019 matrix from base to camera by the transformation 573 00:27:27,019 --> 00:27:29,740 matrix from camera to widget. And the output 574 00:27:29,740 --> 00:27:31,839 is a brand new matrix that tells the robot the 575 00:27:31,839 --> 00:27:34,339 exact pose of the widget relative to its own 576 00:27:34,339 --> 00:27:36,640 base. It knows exactly how to move its arm to 577 00:27:36,640 --> 00:27:39,130 reach the target. But the complexity of that 578 00:27:39,130 --> 00:27:41,750 underlying matrix math depends entirely on whether 579 00:27:41,750 --> 00:27:44,609 the robot is operating in two dimensions or three 580 00:27:44,609 --> 00:27:47,910 dimensions. Cork introduces the 2D math first 581 00:27:47,910 --> 00:27:51,670 to establish the logic. Let's explore that. Imagine 582 00:27:51,670 --> 00:27:54,670 a robot vacuum driving on a perfectly flat floor. 583 00:27:54,910 --> 00:27:57,390 It can't fly. It can't sink. It only exists on 584 00:27:57,390 --> 00:28:00,970 the X and Y axes. In a 2D plane, representing 585 00:28:00,970 --> 00:28:03,630 orientation is beautifully simple. An object 586 00:28:03,630 --> 00:28:06,369 can only spin around one imaginary axis pointing 587 00:28:06,369 --> 00:28:08,990 straight up from the floor. So you only need 588 00:28:08,990 --> 00:28:11,430 a single angle of rotation. Let's call it theta. 589 00:28:12,240 --> 00:28:14,500 Mathematically, Corp. represents this using a 590 00:28:14,500 --> 00:28:17,980 2x2 rotation matrix. This matrix uses the cosine 591 00:28:17,980 --> 00:28:20,480 and sine of theta to map how the object's x and 592 00:28:20,480 --> 00:28:23,839 y axes have rotated relative to the world. This 593 00:28:23,839 --> 00:28:26,500 specific matrix belongs to a group known as SO2, 594 00:28:26,740 --> 00:28:29,039 the special orthogonal group in two dimensions. 595 00:28:29,680 --> 00:28:32,079 It's orthogonal because the axes remain perfectly 596 00:28:32,079 --> 00:28:35,200 perpendicular, and special because it has a determinant 597 00:28:35,200 --> 00:28:38,640 of positive 1, ensuring it represents pure rotation 598 00:28:38,640 --> 00:28:41,839 without stretching or shrinking the object. It's 599 00:28:41,839 --> 00:28:44,480 incredibly elegant. And if you want the full 600 00:28:44,480 --> 00:28:47,220 pose in 2D, adding the positional translation 601 00:28:47,220 --> 00:28:49,819 across the floor to the rotational angle, you 602 00:28:49,819 --> 00:28:52,240 mathematically combine the 2x2 rotation matrix 603 00:28:52,240 --> 00:28:55,559 and the XY translation vector into a larger 3x3 604 00:28:55,559 --> 00:28:58,700 matrix. Known as a homogeneous transformation 605 00:28:58,700 --> 00:29:02,000 matrix, belonging to the Se2 group, the special 606 00:29:02,000 --> 00:29:04,980 Euclidean group in two dimensions. Homogeneous 607 00:29:04,980 --> 00:29:07,839 matrices allow a computer to perform both rotation 608 00:29:07,839 --> 00:29:10,859 and translation simultaneously using a single 609 00:29:10,859 --> 00:29:13,700 matrix multiplication operation. So the 2D math 610 00:29:13,700 --> 00:29:16,519 is tidy and highly efficient, but the physical 611 00:29:16,519 --> 00:29:19,480 universe is not flat. Most robots operate in 612 00:29:19,480 --> 00:29:22,680 3D space, and this is the critical inflection 613 00:29:22,680 --> 00:29:25,380 point of the chapter. Moving from 2D to 3D isn't 614 00:29:25,380 --> 00:29:28,359 just adding one extra variable. The mathematical 615 00:29:28,359 --> 00:29:32,029 complexity explodes exponentially. Because in 616 00:29:32,029 --> 00:29:36,319 2D, rotation is a single angle. In 3D space, 617 00:29:36,700 --> 00:29:39,359 an object can rotate around its x -axis, its 618 00:29:39,359 --> 00:29:42,920 y -axis, and its z -axis independently. The most 619 00:29:42,920 --> 00:29:45,460 intuitive way for human brains to conceptualize 620 00:29:45,460 --> 00:29:49,339 this is through Euler angles, describing 3D rotation 621 00:29:49,339 --> 00:29:51,960 using three distinct parameters. In aerospace, 622 00:29:52,079 --> 00:29:55,019 this is roll, pitch, and yaw. Roll is tilting 623 00:29:55,019 --> 00:29:57,700 the wings. Pitch is pointing the nose up or down. 624 00:29:58,200 --> 00:30:01,250 Yaw is steering left or right. In industrial 625 00:30:01,250 --> 00:30:03,269 robotics, they often use a specific sequence 626 00:30:03,269 --> 00:30:06,789 called ZYZ Euler angles, applying three sequential 627 00:30:06,789 --> 00:30:10,380 rotations. Sounds logical. Three dimensions, 628 00:30:10,460 --> 00:30:13,759 three angles. But Cork explains that utilizing 629 00:30:13,759 --> 00:30:17,119 three parameters to calculate 3D rotation contains 630 00:30:17,119 --> 00:30:20,319 a fatal, inherent mathematical vulnerability. 631 00:30:20,839 --> 00:30:25,150 Gimbal lock. A mathematical singularity. To understand 632 00:30:25,150 --> 00:30:27,589 gimbal lock, visualize a physical gyroscope, 633 00:30:27,670 --> 00:30:30,450 like an inertial measurement unit in early spacecraft. 634 00:30:31,170 --> 00:30:33,690 Picture a heavy spinning brass rotor mounted 635 00:30:33,690 --> 00:30:36,529 inside three concentric metallic rings, or gimbals. 636 00:30:37,250 --> 00:30:39,549 These nested rings let the rotor maintain its 637 00:30:39,549 --> 00:30:42,569 orientation in space, regardless of how the spacecraft 638 00:30:42,569 --> 00:30:45,769 rotates around it. Each ring corresponds to one 639 00:30:45,769 --> 00:30:49,890 axis of rotation. Roll, pitch, and yaw. But here's 640 00:30:49,890 --> 00:30:52,920 the catastrophic mechanical flaw. If the spacecraft 641 00:30:52,920 --> 00:30:55,980 pitches up to a specific extreme angle, exactly 642 00:30:55,980 --> 00:30:59,059 90 degrees, the physical rotation causes the 643 00:30:59,059 --> 00:31:01,259 inner ring to swing around until it perfectly 644 00:31:01,259 --> 00:31:04,299 aligns with the outer ring. And when two of those 645 00:31:04,299 --> 00:31:07,319 rings are parallel and coplanar, they are effectively 646 00:31:07,319 --> 00:31:10,779 pivoting on the exact same axis. You have mechanically 647 00:31:10,779 --> 00:31:13,900 and mathematically lost an entire degree of freedom. 648 00:31:14,500 --> 00:31:17,000 The system is locked. It can no longer measure 649 00:31:17,000 --> 00:31:19,589 rotation in one specific direction. And this 650 00:31:19,589 --> 00:31:21,769 is an abstract theory. During the Apollo Moon 651 00:31:21,769 --> 00:31:24,170 missions, the lunar module relied on a three 652 00:31:24,170 --> 00:31:26,849 -gimbal navigation system. If the astronauts 653 00:31:26,849 --> 00:31:29,369 pitched the lunar module to exactly 90 degrees 654 00:31:29,369 --> 00:31:32,650 during flight, the givils would lock. The navigation 655 00:31:32,650 --> 00:31:35,250 computer would instantly experience a mathematical 656 00:31:35,250 --> 00:31:38,089 singularity. It would completely lose its spatial 657 00:31:38,089 --> 00:31:40,990 orientation, totally unable to calculate which 658 00:31:40,990 --> 00:31:43,650 direction was the Moon and which was deep space. 659 00:31:43,990 --> 00:31:46,849 It was an immense operational hazard. The Apollo 660 00:31:46,849 --> 00:31:49,029 spacecraft actually featured a prominent warning 661 00:31:49,029 --> 00:31:52,250 light on the dashboard, explicitly labeled Gimbal 662 00:31:52,250 --> 00:31:54,809 Lock. If they approached that 90 -degree attitude, 663 00:31:55,150 --> 00:31:58,130 they had to actively intervene, disengage autopilot, 664 00:31:58,529 --> 00:32:00,650 and manually pilot away to prevent the math from 665 00:32:00,650 --> 00:32:03,930 collapsing. So if three numbers cause a catastrophic 666 00:32:03,930 --> 00:32:06,470 collapse, how do modern engineers circumvent 667 00:32:06,470 --> 00:32:09,170 this flaw? The elegant mathematical solution 668 00:32:09,170 --> 00:32:13,000 Cork details is quaternions. By introducing a 669 00:32:13,000 --> 00:32:16,000 fourth parameter, we completely bypass the singularity 670 00:32:16,000 --> 00:32:18,819 of Gimbalock. Quaternions are notoriously difficult 671 00:32:18,819 --> 00:32:21,059 to conceptualize, though, because they require 672 00:32:21,059 --> 00:32:23,700 mapping 3D rotations into a four -dimensional 673 00:32:23,700 --> 00:32:27,059 mathematical space. Totally unintuitive for human 674 00:32:27,059 --> 00:32:30,299 reasoning. Invented by Irish mathematician William 675 00:32:30,299 --> 00:32:33,460 Rowan Hamilton in the 19th century as an extension 676 00:32:33,460 --> 00:32:37,279 of complex numbers, a unit quaternion uses four 677 00:32:37,279 --> 00:32:40,759 numbers, a single scalar part, and a three -dimensional 678 00:32:40,759 --> 00:32:43,549 vector part. How does it actually encode a rotation 679 00:32:43,549 --> 00:32:46,789 without roll, pitch, and yaw? It relies on Euler's 680 00:32:46,789 --> 00:32:49,970 rotation theorem. It defines a single arbitrary 681 00:32:49,970 --> 00:32:52,609 axis of rotation projecting through 3D space, 682 00:32:53,009 --> 00:32:55,190 and then specifies a single angle of rotation 683 00:32:55,190 --> 00:32:58,650 around that specific axis. The scalar and vector 684 00:32:58,650 --> 00:33:02,130 components mathematically encode that axis and 685 00:33:02,130 --> 00:33:04,400 angle. And because the math operates in a four 686 00:33:04,400 --> 00:33:07,400 -dimensional space using four variables, the 687 00:33:07,400 --> 00:33:10,019 axes can never align and collapse. The singularity 688 00:33:10,019 --> 00:33:12,440 simply doesn't exist in that mathematical topology. 689 00:33:12,960 --> 00:33:15,980 Exactly. And beyond solving gimbal lock, quaternions 690 00:33:15,980 --> 00:33:19,619 are computationally beautiful. If a robot uses 691 00:33:19,619 --> 00:33:22,059 a standard rotation matrix for 3D orientation, 692 00:33:22,539 --> 00:33:26,500 the SO3 group, it requires a 3 by 3 matrix. meaning 693 00:33:26,500 --> 00:33:29,039 the processor calculates nine separate numbers 694 00:33:29,039 --> 00:33:31,839 just for rotation. And for the full 3D pose, 695 00:33:32,259 --> 00:33:34,400 position and orientation, you need a homogenous 696 00:33:34,400 --> 00:33:38,259 transformation matrix in SE3, which is a massive 697 00:33:38,259 --> 00:33:41,740 4x4 matrix containing 16 numbers. Multiplying 698 00:33:41,740 --> 00:33:44,799 4x4 matrices requires significant processing 699 00:33:44,799 --> 00:33:48,680 cycles. A Quaternion achieves robust, singularity 700 00:33:48,680 --> 00:33:51,039 -free rotation tracking with only four numbers. 701 00:33:51,599 --> 00:33:54,140 When programming a drone calculating its attitude 702 00:33:54,140 --> 00:33:56,859 a thousand times per second, that reduction in 703 00:33:56,859 --> 00:34:00,220 overhead is massive. Quark also briefly outlines 704 00:34:00,220 --> 00:34:03,220 a practical alternative method for 3D orientation 705 00:34:03,220 --> 00:34:05,819 specifically tailored for industrial robot arms, 706 00:34:06,299 --> 00:34:09,079 the two -vector representation. Right, a perfect 707 00:34:09,079 --> 00:34:11,699 example of applying pragmatic engineering when 708 00:34:11,699 --> 00:34:14,820 complex math isn't necessary. If commanding a 709 00:34:14,820 --> 00:34:17,159 gripper to pick up a box, you don't necessarily 710 00:34:17,159 --> 00:34:20,460 need the abstract 4D space of Quaternions. You 711 00:34:20,460 --> 00:34:23,079 can define the gripper's required orientation 712 00:34:23,079 --> 00:34:26,780 using two distinct physical vectors. First is 713 00:34:26,780 --> 00:34:29,719 the approach vector. Imagine a laser pointer 714 00:34:29,719 --> 00:34:32,820 inside the gripper's wrist, shooting a beam straight 715 00:34:32,820 --> 00:34:35,800 out between the fingers. This dictates the direction 716 00:34:35,800 --> 00:34:37,699 the tool points as it approaches the object. 717 00:34:38,460 --> 00:34:40,539 But just the approach direction is insufficient. 718 00:34:41,199 --> 00:34:43,739 The gripper could point at the box, but be twisted 719 00:34:43,739 --> 00:34:46,739 90 degrees, crushing the sides instead of sliding 720 00:34:46,739 --> 00:34:49,900 under it. To solve that, you add an orientation 721 00:34:49,900 --> 00:34:53,159 vector, a mathematical line drawn perpendicularly 722 00:34:53,159 --> 00:34:56,019 between the two physical fingers. By defining 723 00:34:56,019 --> 00:34:58,659 just those two directional vectors in 3D space, 724 00:34:59,159 --> 00:35:01,280 the controller has all the geometric data to 725 00:35:01,280 --> 00:35:03,920 perfectly align its wrist and safely grasp the 726 00:35:03,920 --> 00:35:07,059 object. Computationally lightweight and physically 727 00:35:07,059 --> 00:35:10,610 intuitive. Brilliant engineering. But... We have 728 00:35:10,610 --> 00:35:13,110 to realize that everything in chapter two, the 729 00:35:13,110 --> 00:35:16,469 matrices, Euler angles, quaternions, describes 730 00:35:16,469 --> 00:35:19,989 a static universe, a freeze frame photograph 731 00:35:19,989 --> 00:35:22,909 of reality. It's the prerequisite geometry, but 732 00:35:22,909 --> 00:35:27,369 it's incomplete because robots move. Which forcefully 733 00:35:27,369 --> 00:35:30,469 transitions us into chapter three, time and motion 734 00:35:30,469 --> 00:35:33,949 dynamics. We are finally unpausing the universe 735 00:35:33,949 --> 00:35:36,670 and introducing the variable of time. When a 736 00:35:36,670 --> 00:35:39,619 pose changes over time. We transition from pure 737 00:35:39,619 --> 00:35:42,920 static geometry into calculus. The mathematical 738 00:35:42,920 --> 00:35:45,760 derivative of pose with respect to time yields 739 00:35:45,760 --> 00:35:49,699 velocity. And just as we divided pose into position 740 00:35:49,699 --> 00:35:53,840 and orientation, we must divide velocity. Translational 741 00:35:53,840 --> 00:35:56,380 velocity is relatively straightforward. It's 742 00:35:56,380 --> 00:35:59,280 simply the rate the origin of a coordinate frame 743 00:35:59,280 --> 00:36:02,159 translates through physical space. How many meters 744 00:36:02,159 --> 00:36:05,360 per second along the x, y, and z axes? A simple 745 00:36:05,360 --> 00:36:08,670 3D vector. Rotational velocity, however, is a 746 00:36:08,670 --> 00:36:12,130 much denser problem. The derivative of a 3D rotation 747 00:36:12,130 --> 00:36:14,550 matrix isn't just a simple rate of change value. 748 00:36:14,710 --> 00:36:17,230 You can't just say degrees per second. Quark 749 00:36:17,230 --> 00:36:20,650 introduces angular velocity. A vector quantity 750 00:36:20,650 --> 00:36:24,070 defining two things simultaneously. The instantaneous 751 00:36:24,070 --> 00:36:26,590 axis around which the frame is rotating at that 752 00:36:26,590 --> 00:36:29,630 precise millisecond, and the magnitude of the 753 00:36:29,630 --> 00:36:32,409 vector defines the absolute rate of spin around 754 00:36:32,409 --> 00:36:35,460 that axis. And Quark represents this using a 755 00:36:35,460 --> 00:36:37,940 highly specific tool called a skew symmetric 756 00:36:37,940 --> 00:36:40,679 matrix. Yes, when you take the derivative of 757 00:36:40,679 --> 00:36:43,500 a rotation matrix, the result is the original 758 00:36:43,500 --> 00:36:46,179 rotation matrix multiplied by a skew symmetric 759 00:36:46,179 --> 00:36:49,420 matrix. Let's demystify this because skew symmetric 760 00:36:49,420 --> 00:36:52,019 sounds intimidating. Mechanically, what is it 761 00:36:52,019 --> 00:36:54,239 actually doing? It's essentially the cross product 762 00:36:54,239 --> 00:36:57,940 operation disguised as a matrix. In vector calculus, 763 00:36:58,639 --> 00:37:00,840 the cross product of an angular velocity vector 764 00:37:00,989 --> 00:37:03,369 and a position vector yields the linear velocity 765 00:37:03,369 --> 00:37:06,650 of that specific point. The skew symmetric matrix 766 00:37:06,650 --> 00:37:08,789 lets a computer perform that cross product using 767 00:37:08,789 --> 00:37:11,809 standard matrix multiplication, mapping rotational 768 00:37:11,809 --> 00:37:14,230 speed into linear speed across a rigid body. 769 00:37:14,829 --> 00:37:17,230 This becomes incredibly vital when we realize 770 00:37:17,230 --> 00:37:20,010 that velocity is entirely relative. It depends 771 00:37:20,010 --> 00:37:22,690 strictly on who is observing the motion. Quark 772 00:37:22,690 --> 00:37:25,329 dives deep into spatial velocities and observers. 773 00:37:25,989 --> 00:37:28,969 The entire mathematical framework relies on perspective. 774 00:37:29,360 --> 00:37:32,199 Let's do a mental experiment. Okay. Imagine a 775 00:37:32,199 --> 00:37:35,179 factory. You have a robotic arm moving rapidly 776 00:37:35,179 --> 00:37:37,719 across a table. Attached to the wrist of that 777 00:37:37,719 --> 00:37:40,940 moving arm is a digital camera. And sitting completely 778 00:37:40,940 --> 00:37:44,380 still on the table is a coffee cup. From the 779 00:37:44,380 --> 00:37:47,460 perspective, the reference frame of the camera 780 00:37:47,460 --> 00:37:49,900 on the wrist, it perceives itself as completely 781 00:37:49,900 --> 00:37:53,199 stationary. To the camera's algorithms, the coffee 782 00:37:53,199 --> 00:37:55,159 cup is the object flying backward through space 783 00:37:55,159 --> 00:37:58,269 at high velocity. But imagine a second camera 784 00:37:58,269 --> 00:38:00,469 bolted to the cinder block wall observing the 785 00:38:00,469 --> 00:38:03,210 scene. From the wall camera's reference frame, 786 00:38:03,409 --> 00:38:05,590 the coffee cup has an absolute velocity of zero. 787 00:38:05,789 --> 00:38:08,590 It's static. And the robotic arm is moving. Both 788 00:38:08,590 --> 00:38:10,329 observations are mathematically valid within 789 00:38:10,329 --> 00:38:13,489 their local reference frames. But how do you 790 00:38:13,489 --> 00:38:15,829 reconcile those conflicting realities inside 791 00:38:15,829 --> 00:38:18,460 the central robot controller? If one sensor says 792 00:38:18,460 --> 00:38:20,539 it's stationary and another says it's moving 793 00:38:20,539 --> 00:38:23,320 at 2 meters per second, the software will crash 794 00:38:23,320 --> 00:38:25,699 unless it can translate between perspectives. 795 00:38:26,280 --> 00:38:28,679 The software needs a mathematical Rosetta Stone, 796 00:38:29,000 --> 00:38:32,380 and that translator is the Jacobian matrix. The 797 00:38:32,380 --> 00:38:34,800 Jacobian is arguably one of the most critical 798 00:38:34,800 --> 00:38:38,340 and complex matrices in all of kinematics. It's 799 00:38:38,340 --> 00:38:41,000 the matrix that haunts the nightmares of robotic 800 00:38:41,000 --> 00:38:44,780 students. How does it physically function? Conceptually, 801 00:38:44,780 --> 00:38:47,730 it's an interaction matrix. composed of partial 802 00:38:47,730 --> 00:38:51,489 derivatives. Imagine a robotic arm with six distinct 803 00:38:51,489 --> 00:38:54,789 rotating joints, a few command joint number three 804 00:38:54,789 --> 00:38:57,789 to rotate slightly. That rotation propagates 805 00:38:57,789 --> 00:39:00,929 down the structure and causes the end effector, 806 00:39:01,130 --> 00:39:04,369 the gripper, to move a specific distance in x, 807 00:39:04,369 --> 00:39:07,630 y, and z space. The Jacobian matrix maps those 808 00:39:07,630 --> 00:39:10,409 individual joint space velocities, how fast each 809 00:39:10,409 --> 00:39:13,809 motor is spinning, directly into task space velocities, 810 00:39:13,849 --> 00:39:17,010 how fast the gripper is moving linearly and rotationally 811 00:39:17,010 --> 00:39:19,929 in the room. And crucially, it operates in reverse. 812 00:39:20,750 --> 00:39:22,309 If you know how fast you want the gripper to 813 00:39:22,309 --> 00:39:24,570 move in a straight line, you multiply that desired 814 00:39:24,570 --> 00:39:27,050 velocity by the inverse of the Jacobian matrix, 815 00:39:27,949 --> 00:39:30,150 and it calculates exactly how fast every single 816 00:39:30,150 --> 00:39:32,489 motor needs to spin to achieve that straight 817 00:39:32,489 --> 00:39:35,469 line motion. It perfectly translates spatial 818 00:39:35,469 --> 00:39:37,969 velocity between complex, interacting coordinate 819 00:39:37,969 --> 00:39:41,010 frames. But the Jacobian has a terrifying flaw. 820 00:39:41,530 --> 00:39:43,690 Because it relies on matrix inversion, it is 821 00:39:43,690 --> 00:39:47,730 susceptible to singularities. If the robot arm 822 00:39:47,730 --> 00:39:50,269 extends completely straight out, reaching its 823 00:39:50,269 --> 00:39:53,289 absolute maximum physical limit, it can no longer 824 00:39:53,289 --> 00:39:55,909 move forward. Mathematically, the determinant 825 00:39:55,909 --> 00:39:58,929 of the Jacobian matrix drops to zero. And when 826 00:39:58,929 --> 00:40:01,530 a matrix determinant is zero, you cannot calculate 827 00:40:01,530 --> 00:40:04,130 its inverse. The math literally divides by zero 828 00:40:04,130 --> 00:40:06,690 and explodes. The controller loses the ability 829 00:40:06,690 --> 00:40:09,269 to calculate joint speeds and the physical arm 830 00:40:09,269 --> 00:40:11,550 gets stuck until a human physically moves it 831 00:40:11,550 --> 00:40:14,230 out of the singularity. This is why calculating 832 00:40:14,230 --> 00:40:17,230 and avoiding Jacobian singularities is a massive 833 00:40:17,230 --> 00:40:20,480 focus of trajectory planning. Now, everything 834 00:40:20,480 --> 00:40:23,039 we've covered regarding velocities and the Jacobian 835 00:40:23,039 --> 00:40:26,079 falls under kinematics. strictly the geometry 836 00:40:26,079 --> 00:40:28,860 of motion. It describes how things are moving, 837 00:40:29,260 --> 00:40:32,780 but willfully ignores why they are moving. To 838 00:40:32,780 --> 00:40:35,519 understand the why we transitioned into dynamics. 839 00:40:36,219 --> 00:40:39,219 Dynamics introduces mass, physical inertia, forces, 840 00:40:39,460 --> 00:40:42,199 and torques into the mathematical universe. Dynamics 841 00:40:42,199 --> 00:40:45,239 is the physics engine that dictates the kinematics. 842 00:40:45,650 --> 00:40:48,170 Cork brings in the fundamental laws of classical 843 00:40:48,170 --> 00:40:50,809 physics here. For a translational motion, moving 844 00:40:50,809 --> 00:40:53,389 in a straight line, the robot must obey Sir Isaac 845 00:40:53,389 --> 00:40:56,329 Newton's second law. Force equals mass times 846 00:40:56,329 --> 00:40:58,710 acceleration. If a motor applies linear force 847 00:40:58,710 --> 00:41:02,250 to a mass, it generates acceleration. But robotic 848 00:41:02,250 --> 00:41:05,750 arms primarily rotate. For rotational motion 849 00:41:05,750 --> 00:41:08,590 in 3D space, the math escalates to Euler's equations 850 00:41:08,590 --> 00:41:11,570 of motion. And Euler's equations introduce a 851 00:41:11,570 --> 00:41:14,369 beast of a matrix called the rotational inertia 852 00:41:14,369 --> 00:41:17,960 tensor. The inertia tensor is a 3x3 matrix that 853 00:41:17,960 --> 00:41:20,659 mathematically describes precisely how the physical 854 00:41:20,659 --> 00:41:23,000 mass of a robotic link is distributed around 855 00:41:23,000 --> 00:41:25,800 its center of gravity. It's the figure skater 856 00:41:25,800 --> 00:41:28,539 physics. When a figure skater spins on the ice 857 00:41:28,539 --> 00:41:30,940 with their arms extended wide, their mass is 858 00:41:30,940 --> 00:41:33,559 distributed far from the axis of rotation, creating 859 00:41:33,559 --> 00:41:36,699 high rotational inertia. They spin slowly. When 860 00:41:36,699 --> 00:41:38,599 they pull their arms in tight to their chest, 861 00:41:39,000 --> 00:41:41,179 the inertia tensor changes drastically and they 862 00:41:41,179 --> 00:41:44,429 instantly spin much faster. A robotic arm experiences 863 00:41:44,429 --> 00:41:47,369 the exact same physics. An arm fully extended 864 00:41:47,369 --> 00:41:49,769 horizontally with a heavy payload in its gripper 865 00:41:49,769 --> 00:41:52,130 has a drastically different inertia tensor than 866 00:41:52,130 --> 00:41:54,969 the same arm folded compactly against its space. 867 00:41:55,650 --> 00:41:58,150 Therefore, the control software must constantly 868 00:41:58,150 --> 00:42:01,530 recalculate the inertia tensor to determine exactly 869 00:42:01,530 --> 00:42:04,550 how much electrical torque the motors must exert 870 00:42:04,550 --> 00:42:07,449 to rotate the arm at the desired speed. Which 871 00:42:07,449 --> 00:42:09,650 introduces an incredibly critical distinction 872 00:42:09,650 --> 00:42:12,739 Quark makes regarding the very nature of the 873 00:42:12,739 --> 00:42:14,940 reference frames we use to calculate this physics. 874 00:42:15,440 --> 00:42:17,900 You must know with absolute certainty whether 875 00:42:17,900 --> 00:42:21,639 your coordinate frame is inertial or non -inertial. 876 00:42:22,199 --> 00:42:24,980 This is a vital physical distinction. And the 877 00:42:24,980 --> 00:42:26,699 inertial reference frame is strictly defined 878 00:42:26,699 --> 00:42:29,480 as a frame that is not accelerating and is not 879 00:42:29,480 --> 00:42:32,440 rotating. It's essentially anchored to a fixed 880 00:42:32,440 --> 00:42:34,880 point in the universe or moving at a perfectly 881 00:42:34,880 --> 00:42:37,599 constant velocity in a perfectly straight line. 882 00:42:38,179 --> 00:42:40,150 Why does that matter? Because Newton's laws of 883 00:42:40,150 --> 00:42:42,650 motion and all the dynamic calculations based 884 00:42:42,650 --> 00:42:45,949 on them only function correctly within an inertial 885 00:42:45,949 --> 00:42:48,690 frame. If your reference frame is accelerating, 886 00:42:49,409 --> 00:42:52,639 physics appears to break down. We all experience 887 00:42:52,639 --> 00:42:55,780 this phenomenon intuitively. Think about sitting 888 00:42:55,780 --> 00:42:58,239 in the passenger seat of a high -performance 889 00:42:58,239 --> 00:43:00,500 sports car stopped at a red light. The light 890 00:43:00,500 --> 00:43:03,380 turns green, the driver forcefully floors the 891 00:43:03,380 --> 00:43:06,159 accelerator pedal, and the vehicle violently 892 00:43:06,159 --> 00:43:09,099 accelerates forward. What is your physical experience? 893 00:43:09,579 --> 00:43:11,440 You feel a heavy force pushing you backward, 894 00:43:11,460 --> 00:43:13,539 pinning your spine against the leather seat. 895 00:43:14,000 --> 00:43:16,940 Exactly. But here is the bizarre reality of physics. 896 00:43:17,320 --> 00:43:20,079 There is absolutely no physical force pushing 897 00:43:20,079 --> 00:43:23,159 you backward. Nothing touched your chest. The 898 00:43:23,159 --> 00:43:25,659 only actual physical force occurring in the universe 899 00:43:25,659 --> 00:43:28,639 in that moment is the structure of the car seat 900 00:43:28,639 --> 00:43:32,039 pushing you forward. However, because your perspective 901 00:43:32,039 --> 00:43:34,619 is located inside the accelerating cabin of the 902 00:43:34,619 --> 00:43:38,409 car, Your coordinate frame is non -inertial. 903 00:43:38,550 --> 00:43:40,630 From within a non -inertial frame, you experience 904 00:43:40,630 --> 00:43:44,250 what physicists call a fictitious force. It feels 905 00:43:44,250 --> 00:43:46,710 completely real, but it's an illusion created 906 00:43:46,710 --> 00:43:48,670 by the acceleration of your reference frame. 907 00:43:49,269 --> 00:43:51,630 Robotic systems experience these fictitious forces 908 00:43:51,630 --> 00:43:54,769 constantly. Consider a camera rigidly mounted 909 00:43:54,769 --> 00:43:57,630 to a high -speed racing drone, executing sharp, 910 00:43:57,730 --> 00:44:00,670 high -G maneuvers. As the drone accelerates and 911 00:44:00,670 --> 00:44:03,150 rotates violently, the camera's coordinate frame 912 00:44:03,150 --> 00:44:05,579 becomes highly non -inertial. It experiences 913 00:44:05,579 --> 00:44:09,159 extreme fictitious forces, specifically the Coriolis 914 00:44:09,159 --> 00:44:12,079 effect and centrifugal force. If the software 915 00:44:12,079 --> 00:44:14,460 algorithms processing the navigation data don't 916 00:44:14,460 --> 00:44:16,840 mathematically compensate for the fact that their 917 00:44:16,840 --> 00:44:19,400 reference frame is accelerating, the drone's 918 00:44:19,400 --> 00:44:21,860 internal model of reality will fail catastrophically 919 00:44:21,860 --> 00:44:24,500 and it will crash. So we have established the 920 00:44:24,500 --> 00:44:27,380 geometry of motion and we comprehend the dynamic 921 00:44:27,380 --> 00:44:31,119 physics of forces causing acceleration. But a 922 00:44:31,119 --> 00:44:33,739 crucial operational question remains. How do 923 00:44:33,739 --> 00:44:36,679 we actually command a robot to execute a movement? 924 00:44:37,380 --> 00:44:40,219 We can't simply feed a robot arm a start coordinate 925 00:44:40,219 --> 00:44:42,159 and an end coordinate and expect it to figure 926 00:44:42,159 --> 00:44:45,179 it out. If you just say, go from point A to point 927 00:44:45,179 --> 00:44:48,360 B, the motors will instantly apply maximum torque. 928 00:44:48,739 --> 00:44:51,460 The arm will swing wildly, overshoot the target, 929 00:44:51,559 --> 00:44:53,960 and likely smash into the workbench. We need 930 00:44:53,960 --> 00:44:56,559 controlled motion. This brings us into the final 931 00:44:56,559 --> 00:44:59,920 section of our deep dive, trajectories and navigation, 932 00:45:00,579 --> 00:45:03,750 the concluding half of chapter three. The objective 933 00:45:03,750 --> 00:45:06,829 is to command the robot to traverse space elegantly 934 00:45:06,829 --> 00:45:09,429 without destroying its own gearboxes. To achieve 935 00:45:09,429 --> 00:45:12,289 this, Cork relies on a concept called configuration 936 00:45:12,289 --> 00:45:15,769 space, or C -space. Imagine a complex industrial 937 00:45:15,769 --> 00:45:18,530 arm with six rotating joints. Trying to plot 938 00:45:18,530 --> 00:45:20,889 a path for the physical geometry of that arm 939 00:45:20,889 --> 00:45:23,710 around obstacles in 3D space is incredibly difficult. 940 00:45:24,329 --> 00:45:27,409 C -space simplifies this mathematically. Instead 941 00:45:27,409 --> 00:45:30,550 of plotting the 3D arm, you map the current angle 942 00:45:30,550 --> 00:45:33,090 of all six joints into a single coordinate in 943 00:45:33,090 --> 00:45:35,969 a six -dimensional mathematical space. Every 944 00:45:35,969 --> 00:45:38,030 unique posture the robot can assume corresponds 945 00:45:38,030 --> 00:45:41,130 to exactly one point in this C -space. And the 946 00:45:41,130 --> 00:45:43,550 goal of trajectory planning is simply to draw 947 00:45:43,550 --> 00:45:47,389 a line, a time -varying sequence of poses, through 948 00:45:47,389 --> 00:45:50,050 that C -space, connecting your start point to 949 00:45:50,050 --> 00:45:52,909 your desired end point while avoiding areas that 950 00:45:52,909 --> 00:45:54,929 represent physical collisions with obstacles. 951 00:45:55,280 --> 00:45:58,340 But it can't be just any arbitrary line. To prevent 952 00:45:58,340 --> 00:46:01,019 mechanical damage, the generated trajectory must 953 00:46:01,019 --> 00:46:04,260 possess continuous, smooth mathematical derivatives. 954 00:46:05,099 --> 00:46:07,719 Let's unpack the necessity of those smooth derivatives. 955 00:46:08,460 --> 00:46:11,260 First, the position trajectory must be continuous. 956 00:46:11,639 --> 00:46:14,280 The robot cannot mathematically teleport from 957 00:46:14,280 --> 00:46:17,000 A to B. It must traverse the intervening space. 958 00:46:17,800 --> 00:46:20,599 Second, the velocity profile must be continuous. 959 00:46:20,940 --> 00:46:23,760 You cannot command a joint to instantaneously 960 00:46:23,760 --> 00:46:26,820 accelerate from zero to 50 rpm in zero seconds 961 00:46:26,820 --> 00:46:30,400 because Newton's laws dictate that requires infinite 962 00:46:30,400 --> 00:46:33,679 acceleration, which requires infinite force. 963 00:46:34,420 --> 00:46:36,460 Third, and this is crucial for the physical longevity 964 00:46:36,460 --> 00:46:39,300 of the robot, the acceleration profile should 965 00:46:39,300 --> 00:46:42,579 ideally be continuous. A sudden instantaneous 966 00:46:42,579 --> 00:46:45,280 spike in acceleration is mathematically defined 967 00:46:45,280 --> 00:46:48,500 as jerk. Jerk is the violent shuddering you see 968 00:46:48,500 --> 00:46:51,340 in poorly programmed machinery. It destroys gear 969 00:46:51,340 --> 00:46:54,139 teeth and burns out motor windings. To eliminate 970 00:46:54,139 --> 00:46:57,019 Jerk, roboticists utilize complex mathematical 971 00:46:57,019 --> 00:46:59,900 polynomials, often quintic fifth -order polynomials, 972 00:47:00,260 --> 00:47:02,539 to define the trajectory. These mathematical 973 00:47:02,539 --> 00:47:05,119 curves guarantee that the robot will gradually 974 00:47:05,119 --> 00:47:07,880 ramp up its acceleration, achieve a smooth cruising 975 00:47:07,880 --> 00:47:10,659 velocity, and elegantly decelerate as it approaches 976 00:47:10,659 --> 00:47:13,940 the target. But while generating smooth polynomials 977 00:47:13,940 --> 00:47:17,239 for positional translation is relatively straightforward, 978 00:47:18,079 --> 00:47:20,440 interpolating orientation across a trajectory 979 00:47:20,440 --> 00:47:23,599 forces us to confront the bizarre topology of 980 00:47:23,599 --> 00:47:26,480 3D rotational math once again. Yes, this is a 981 00:47:26,480 --> 00:47:29,639 fascinating mathematical trap. If you are commanding 982 00:47:29,639 --> 00:47:32,400 a robotic wrist to rotate from orientation A 983 00:47:32,400 --> 00:47:35,139 to orientation B, you want it to take the shortest, 984 00:47:35,159 --> 00:47:38,039 most efficient path. But in rotational space, 985 00:47:38,059 --> 00:47:40,460 there are always two paths to reach any given 986 00:47:40,460 --> 00:47:43,530 destination. If you were standing still and want 987 00:47:43,530 --> 00:47:45,869 to face an object positioned 10 degrees to your 988 00:47:45,869 --> 00:47:48,869 right, the most efficient physical path is to 989 00:47:48,869 --> 00:47:51,750 simply pivot 10 degrees to the right. However, 990 00:47:52,090 --> 00:47:54,389 you could alternatively pivot 350 degrees to 991 00:47:54,389 --> 00:47:56,969 your left. Both movements place you in the exact 992 00:47:56,969 --> 00:47:59,489 same final orientation, but one is vastly less 993 00:47:59,489 --> 00:48:02,570 efficient. Earlier we praised quaternions for 994 00:48:02,570 --> 00:48:04,610 gracefully solving the Gimbalock singularity. 995 00:48:04,719 --> 00:48:07,639 And Quirk explains that to smoothly transition 996 00:48:07,639 --> 00:48:10,360 a robot's orientation between two quaternions, 997 00:48:10,880 --> 00:48:13,699 we utilize a mathematical operation called SLERP. 998 00:48:14,159 --> 00:48:16,840 SLERP stands for Spherical Linear Interpolation. 999 00:48:17,260 --> 00:48:19,599 It calculates the smoothest constant velocity 1000 00:48:19,599 --> 00:48:21,760 arc between two points on the surface of a four 1001 00:48:21,760 --> 00:48:24,679 -dimensional sphere. But quaternions harbor a 1002 00:48:24,679 --> 00:48:27,699 deeply weird topological quirk known as double 1003 00:48:27,699 --> 00:48:30,179 cover. Because they exist on a four -dimensional 1004 00:48:30,179 --> 00:48:33,840 hypersphere, any single physical 3D orientation 1005 00:48:34,170 --> 00:48:37,250 actually maps to two distinct quaternions located 1006 00:48:37,250 --> 00:48:39,510 on the exact opposite sides of the sphere from 1007 00:48:39,510 --> 00:48:43,130 each other. The quaternion Q and the quaternion 1008 00:48:43,130 --> 00:48:45,849 negative Q represent the exact same physical 1009 00:48:45,849 --> 00:48:49,110 attitude of the robot. This creates a dangerous 1010 00:48:49,110 --> 00:48:52,409 pitfall during trajectory generation. If a programmer 1011 00:48:52,409 --> 00:48:55,010 is not intensely careful with their Python implementation, 1012 00:48:55,750 --> 00:48:58,389 they can command the robot to use SLERP to transition 1013 00:48:58,389 --> 00:49:00,829 from its current quaternion pose to a target 1014 00:49:00,829 --> 00:49:03,800 quaternion pose. The SLERP algorithm will execute 1015 00:49:03,800 --> 00:49:06,739 perfectly, calculating a mathematically flawless, 1016 00:49:07,039 --> 00:49:10,179 buttersmooth trajectory. But because it mathematically 1017 00:49:10,179 --> 00:49:12,480 locked onto the quaternion on the far side of 1018 00:49:12,480 --> 00:49:14,699 the sphere, the algorithm forces the physical 1019 00:49:14,699 --> 00:49:17,360 robot to take the long way around. You command 1020 00:49:17,360 --> 00:49:19,739 a minor 10 -degree wrist adjustment, and the 1021 00:49:19,739 --> 00:49:22,559 software violently swings the robotic arm 350 1022 00:49:22,559 --> 00:49:25,019 degrees in the opposite direction, tearing the 1023 00:49:25,019 --> 00:49:26,679 pneumatic hoses out of the wall and smashing 1024 00:49:26,679 --> 00:49:29,179 the payload, merely to arrive at the correct 1025 00:49:29,179 --> 00:49:32,389 final orientation. To prevent this specific catastrophe, 1026 00:49:32,929 --> 00:49:35,269 the control software must explicitly perform 1027 00:49:35,269 --> 00:49:38,070 a dot product mathematical check between the 1028 00:49:38,070 --> 00:49:40,710 starting and ending quaternions before executing 1029 00:49:40,710 --> 00:49:43,070 the move. If the math indicates it is about to 1030 00:49:43,070 --> 00:49:46,110 take the long path, the software manually negates 1031 00:49:46,110 --> 00:49:48,690 the target quaternion, effectively flipping it 1032 00:49:48,690 --> 00:49:51,530 to the near side of the sphere, forcing the algorithm 1033 00:49:51,530 --> 00:49:53,929 to take the absolute shortest physical path. 1034 00:49:54,539 --> 00:49:57,639 It's a stunning example of how abstract, seemingly 1035 00:49:57,639 --> 00:50:00,500 esoteric math concepts, like the topology of 1036 00:50:00,500 --> 00:50:03,360 a four -dimensional sphere, have immediate, highly 1037 00:50:03,360 --> 00:50:05,420 destructive consequences in the real world of 1038 00:50:05,420 --> 00:50:08,360 robotics. Now, Corse brilliantly ties all this 1039 00:50:08,360 --> 00:50:12,179 abstract math, the quaternions, angular velocity 1040 00:50:12,179 --> 00:50:15,039 derivatives, complex reference frames, into a 1041 00:50:15,039 --> 00:50:17,559 highly practical real -world application at the 1042 00:50:17,559 --> 00:50:20,539 conclusion of chapter 3, the mathematics of inertial 1043 00:50:20,539 --> 00:50:23,670 navigation. The foundational question of inertial 1044 00:50:23,670 --> 00:50:26,469 navigation is this. How does a robotic system 1045 00:50:26,469 --> 00:50:29,250 determine its precise location and orientation 1046 00:50:29,250 --> 00:50:32,530 when it is completely isolated? It's operating 1047 00:50:32,530 --> 00:50:35,429 in total darkness, its vision sensors are useless, 1048 00:50:35,769 --> 00:50:38,289 and it is deep underground where GPS signals 1049 00:50:38,289 --> 00:50:41,389 cannot penetrate. It navigates by sensing its 1050 00:50:41,389 --> 00:50:44,449 own internal physics and meticulously tracking 1051 00:50:44,449 --> 00:50:47,820 its history. The robot utilizes highly sensitive 1052 00:50:47,820 --> 00:50:50,940 onboard sensors, specifically gyroscopes, to 1053 00:50:50,940 --> 00:50:53,719 measure its instantaneous angular velocity. Modern 1054 00:50:53,719 --> 00:50:57,039 systems use microelectromechanical systems, or 1055 00:50:57,039 --> 00:51:00,179 MEMS gyroscopes. These are microscopic vibrating 1056 00:51:00,179 --> 00:51:02,920 structures etched into silicone chips. As the 1057 00:51:02,920 --> 00:51:07,000 robot rotates, the Coriolis effect causes microscopic 1058 00:51:07,000 --> 00:51:09,480 shifts in the vibration, which the chip measures 1059 00:51:09,480 --> 00:51:12,420 as angular velocity in the X, Y, and Z axes. 1060 00:51:12,780 --> 00:51:15,260 If the robot controller knows its exact starting 1061 00:51:15,260 --> 00:51:18,260 orientation when it powered on, and the gyroscope 1062 00:51:18,260 --> 00:51:20,639 streams data detailing exactly how many degrees 1063 00:51:20,639 --> 00:51:23,340 per second it is rotating at every millisecond, 1064 00:51:23,960 --> 00:51:25,920 the software can utilize calculus to integrate 1065 00:51:25,920 --> 00:51:28,820 that angular velocity over time. By constantly 1066 00:51:28,820 --> 00:51:32,059 adding up those tiny rotational changes, it constantly 1067 00:51:32,059 --> 00:51:34,559 updates its mathematical estimate of its current 1068 00:51:34,559 --> 00:51:37,639 orientation. It's dead reckoning based entirely 1069 00:51:37,639 --> 00:51:40,480 on the physics of rotation. And here is a realization 1070 00:51:40,480 --> 00:51:43,130 that is absolutely profoundly mind -blowing. 1071 00:51:44,210 --> 00:51:46,670 Nature engineered this exact inertial navigation 1072 00:51:46,670 --> 00:51:50,030 system hundreds of millions of years before humanity 1073 00:51:50,030 --> 00:51:53,030 invented the silicon gyroscope. The biological 1074 00:51:53,030 --> 00:51:55,409 equivalent is the vestibular system, located 1075 00:51:55,409 --> 00:51:57,909 deep within the inner ear of almost all vertebrates, 1076 00:51:58,070 --> 00:52:00,449 including humans. Inside your skull, you have 1077 00:52:00,449 --> 00:52:04,050 three tiny fluid -filled tubes called the semicircular 1078 00:52:04,050 --> 00:52:07,349 canals. And evolutionary biology arranged them 1079 00:52:07,349 --> 00:52:10,760 completely orthogonally to each other. One canal 1080 00:52:10,760 --> 00:52:13,980 is dedicated to the x -axis, one to the y -axis, 1081 00:52:14,199 --> 00:52:17,440 and one to the z -axis. They are structurally 1082 00:52:17,440 --> 00:52:21,039 and functionally identical to a triaxial robotic 1083 00:52:21,039 --> 00:52:24,039 gyroscope. When you turn your head, the fluid 1084 00:52:24,039 --> 00:52:26,800 inside the canal sloshes against microscopic 1085 00:52:26,800 --> 00:52:30,340 hair cells. Those cells generate electrical impulses, 1086 00:52:30,420 --> 00:52:33,139 angular velocity data, and transmit them down 1087 00:52:33,139 --> 00:52:35,820 the vestibular nerve to your brainstem, which 1088 00:52:35,820 --> 00:52:38,639 integrates the data to update your physical orientation 1089 00:52:38,639 --> 00:52:42,460 in space. It's an incredible convergence. Robotic 1090 00:52:42,460 --> 00:52:45,300 engineering mathematically reinvented the biological 1091 00:52:45,300 --> 00:52:48,849 inner ear. However, Cork utilizes this section 1092 00:52:48,849 --> 00:52:51,849 to issue a stark warning regarding the mathematical 1093 00:52:51,849 --> 00:52:58,639 limitations of inertial navigation. When a computer 1094 00:52:58,639 --> 00:53:01,179 utilizes calculus to integrate a continuous stream 1095 00:53:01,179 --> 00:53:03,739 of data over time, it is essentially calculating 1096 00:53:03,739 --> 00:53:07,619 a massive running sum. And physical sensors are 1097 00:53:07,619 --> 00:53:11,159 never perfect. Exactly the problem. Imagine a 1098 00:53:11,159 --> 00:53:13,320 cheap silicon gyroscope mounted on a stationary 1099 00:53:13,320 --> 00:53:16,739 robot. Because of thermal noise or manufacturing 1100 00:53:16,739 --> 00:53:19,079 imperfections, the sensor might output a reading 1101 00:53:19,079 --> 00:53:22,820 that says it is rotating at 0 .001 degrees per 1102 00:53:22,820 --> 00:53:25,840 second, even though it is sitting perfectly still 1103 00:53:25,840 --> 00:53:29,219 on a desk. That tiny microscopic error is called 1104 00:53:29,219 --> 00:53:32,059 bias instability because the software is constantly 1105 00:53:32,059 --> 00:53:34,679 integrating, constantly adding that data up. 1106 00:53:35,039 --> 00:53:37,519 That microscopic error accumulates. Second by 1107 00:53:37,519 --> 00:53:40,679 second, the error grows. Minute by minute, the 1108 00:53:40,679 --> 00:53:43,340 robot's internal mathematical model of reality 1109 00:53:43,340 --> 00:53:47,139 slowly drifts away from physical reality. After 1110 00:53:47,139 --> 00:53:49,960 an hour of relying purely on a cheap gyroscope, 1111 00:53:50,340 --> 00:53:52,639 the robot's internal matrix might dictate that 1112 00:53:52,639 --> 00:53:55,800 it is facing true north when in physical reality, 1113 00:53:55,949 --> 00:53:59,349 it is facing due west. This integration drift 1114 00:53:59,349 --> 00:54:02,550 is an unavoidable physical reality of pure inertial 1115 00:54:02,550 --> 00:54:05,369 navigation over long durations. To solve it, 1116 00:54:05,610 --> 00:54:08,469 roboticists must employ a technique called sensor 1117 00:54:08,469 --> 00:54:11,710 fusion. The robot must eventually cross -reference 1118 00:54:11,710 --> 00:54:14,750 its internal math against an external truth. 1119 00:54:15,449 --> 00:54:18,690 It runs complex algorithms like a Kalman filter 1120 00:54:18,690 --> 00:54:21,750 to combine the fast, smooth data from the inertial 1121 00:54:21,750 --> 00:54:25,159 gyroscopes with the slower, absolute data from 1122 00:54:25,159 --> 00:54:28,380 an extroceptive sensor. Perhaps taking a snapshot 1123 00:54:28,380 --> 00:54:31,800 with a camera or catching a brief GPS ping. It 1124 00:54:31,800 --> 00:54:34,159 uses the external data to mathematically correct 1125 00:54:34,159 --> 00:54:36,639 the drift, resetting its internal orientation 1126 00:54:36,639 --> 00:54:40,159 to reality. It's a constant dynamic loop of internal 1127 00:54:40,159 --> 00:54:43,860 estimation and external correction. So as we 1128 00:54:43,860 --> 00:54:46,880 step back and review the massive volume of material 1129 00:54:46,880 --> 00:54:49,380 we've covered in part one of this margin alias 1130 00:54:49,380 --> 00:54:52,480 series. tracing the lineage from an 18th century 1131 00:54:52,480 --> 00:54:55,800 clockwork duck driven by brass cams, through 1132 00:54:55,800 --> 00:54:58,780 the ethical dilemmas of shared control, and deep 1133 00:54:58,780 --> 00:55:01,880 into the four -dimensional topology of quaternions 1134 00:55:01,880 --> 00:55:05,570 and the calculus of inertial drift. What is the 1135 00:55:05,570 --> 00:55:07,989 ultimate synthesis? The foundational takeaway 1136 00:55:07,989 --> 00:55:10,789 is that a robotic system is significantly more 1137 00:55:10,789 --> 00:55:13,730 than its mechanical hardware. The steel linkages, 1138 00:55:13,849 --> 00:55:16,750 the lithium batteries, the electric motors, those 1139 00:55:16,750 --> 00:55:19,960 are merely the physical vessel. At its core, 1140 00:55:20,039 --> 00:55:22,599 to successfully interact with the universe, a 1141 00:55:22,599 --> 00:55:25,780 robot is a continuously executing, highly complex 1142 00:55:25,780 --> 00:55:28,480 mathematical transformation. It is a machine 1143 00:55:28,480 --> 00:55:31,139 that ingests chaotic physical reality through 1144 00:55:31,139 --> 00:55:34,920 sensors, maps that data into vast multi -dimensional 1145 00:55:34,920 --> 00:55:38,710 matrices, translates those matrices across space 1146 00:55:38,710 --> 00:55:42,010 and time using rigorous calculus, and finally 1147 00:55:42,010 --> 00:55:44,929 outputs that math back into the physical world 1148 00:55:44,929 --> 00:55:48,409 as physical force. It is pure geometry made kinetic. 1149 00:55:48,670 --> 00:55:51,250 Which leaves us with a deeply compelling philosophical 1150 00:55:51,250 --> 00:55:53,670 query to ponder, building directly upon Cork's 1151 00:55:53,670 --> 00:55:55,570 connections between biology and engineering. 1152 00:55:56,210 --> 00:55:58,369 If the biological structures of the human inner 1153 00:55:58,369 --> 00:56:01,449 ear operate on the exact same orthogonal triaxial 1154 00:56:01,449 --> 00:56:04,469 principles as a silicon gyroscope array, and 1155 00:56:04,469 --> 00:56:06,650 if the human visual cortex evolved to process 1156 00:56:06,650 --> 00:56:10,750 photons and solve the exact same 3D to 2D projective 1157 00:56:10,750 --> 00:56:13,329 geometry flattening problem that confounds robotic 1158 00:56:13,329 --> 00:56:16,309 camera arrays, At what specific threshold of 1159 00:56:16,309 --> 00:56:18,949 complexity does a biological navigation system 1160 00:56:18,949 --> 00:56:22,429 become mathematically and computationally indistinguishable 1161 00:56:22,429 --> 00:56:24,789 from an engineered robotic navigation system? 1162 00:56:25,150 --> 00:56:28,050 Are we, in the most fundamental mechanical sense, 1163 00:56:28,769 --> 00:56:31,690 just highly complex fleshy coordinate frames, 1164 00:56:32,230 --> 00:56:34,550 navigating our way through a massive universal 1165 00:56:34,550 --> 00:56:38,030 pose graph? It is a profound, unsettling puzzle 1166 00:56:38,030 --> 00:56:40,429 to consider the next time you effortlessly reach 1167 00:56:40,429 --> 00:56:43,050 out and grab a cup of coffee, a testament to 1168 00:56:43,050 --> 00:56:45,250 the unseen mathematics operating inside your 1169 00:56:45,250 --> 00:56:48,550 own skull. A great point to end on. That completes 1170 00:56:48,550 --> 00:56:50,929 our extensive foundation. We have successfully 1171 00:56:50,929 --> 00:56:53,050 wrapped up part one of our deep dive into Peter 1172 00:56:53,050 --> 00:56:56,289 Cork's robotics, vision, and control. We have 1173 00:56:56,289 --> 00:56:58,869 established the rigorous laws of physics, time, 1174 00:56:58,989 --> 00:57:01,510 and geometry, and we have mathematically encoded 1175 00:57:01,510 --> 00:57:05,210 them. The universe is built. Next time, we are 1176 00:57:05,210 --> 00:57:07,409 going to place that mathematics into motion across 1177 00:57:07,409 --> 00:57:10,250 the ground. Prepare for part two of five in the 1178 00:57:10,250 --> 00:57:13,070 series, Mobile Robotics. We will see you then.