1 00:00:00,040 --> 00:00:02,160 Speaker 1: Close your eyes for a second, seriously, just for a moment. 2 00:00:02,200 --> 00:00:04,719 I want you to picture a scene. Okay, The lights 3 00:00:04,759 --> 00:00:08,199 go down in a massive arena. The crowd goes dead, silent, 4 00:00:08,359 --> 00:00:12,240 just holding their breath. A figure steps into the octagon. 5 00:00:12,679 --> 00:00:15,599 It has the silhouette of a middleweight fighter, you know, 6 00:00:15,839 --> 00:00:19,640 about one hundred and sixty five pounds, broad shoulders, looks 7 00:00:19,679 --> 00:00:20,559 ready to throw. 8 00:00:20,399 --> 00:00:22,320 Speaker 2: Down, right, a classic fighter's build. 9 00:00:22,480 --> 00:00:25,800 Speaker 1: But then the spotlight hits it and you realize something 10 00:00:25,879 --> 00:00:29,039 is off. This thing isn't sweating. It doesn't have skin, 11 00:00:29,600 --> 00:00:33,880 it's built from aviation grade aluminum. It doesn't have muscles, 12 00:00:33,960 --> 00:00:37,880 it has high torque actuators, and it definitely doesn't have fear. 13 00:00:38,119 --> 00:00:39,520 Speaker 2: It just has a prime directive. 14 00:00:39,960 --> 00:00:43,280 Speaker 1: What's that stay standing exactly? The bell rings and this 15 00:00:43,320 --> 00:00:46,000 thing moves with a speed that just shouldn't be possible 16 00:00:46,000 --> 00:00:48,079 for metal. It sounds like a scene from a sci 17 00:00:48,119 --> 00:00:48,920 fi blockbuster. 18 00:00:49,240 --> 00:00:54,159 Speaker 2: Yeah, absolutely, somebody'd watch it go Wow, the CGI's amazing, right. 19 00:00:53,920 --> 00:00:57,039 Speaker 1: Cool graphics. But here's the thread we are pulling today. 20 00:00:57,479 --> 00:01:00,000 This isn't a movie. It isn't CGI. It is happy 21 00:01:00,439 --> 00:01:05,000 right now, A literal robot fight club has been announced 22 00:01:05,480 --> 00:01:11,120 and the hardware involved is. Honestly, it's terrifyingly capable. 23 00:01:11,239 --> 00:01:13,640 Speaker 2: It really is a moment where the boundary between special 24 00:01:13,640 --> 00:01:17,120 effects and breaking news has completely dissolved. Yea, we aren't 25 00:01:17,120 --> 00:01:20,840 looking at animation anymore. We're looking at raw physics executed 26 00:01:20,840 --> 00:01:21,920 by autonomous systems. 27 00:01:22,000 --> 00:01:24,959 Speaker 1: Welcome to thrilling Threads. I'm your host, And today we 28 00:01:25,000 --> 00:01:27,159 are looking at a week where the world of robotics 29 00:01:27,239 --> 00:01:30,319 didn't just take a step forward, it did a backflip, 30 00:01:30,599 --> 00:01:33,840 learned to skateboard, and arguably started training for war. 31 00:01:34,000 --> 00:01:36,599 Speaker 2: It has been a breathless few days. We're seeing a 32 00:01:36,599 --> 00:01:42,439 convergence of durability, agility, and perhaps most unervingly, personality in 33 00:01:42,480 --> 00:01:43,079 these machines. 34 00:01:43,159 --> 00:01:45,599 Speaker 1: We have a massive stack of reports to get through today, 35 00:01:45,640 --> 00:01:48,400 we're primarily drawing from the incredible reporting done by the 36 00:01:48,400 --> 00:01:49,200 AI Nexus. 37 00:01:49,239 --> 00:01:51,200 Speaker 2: They've been on top of this beat like no one. 38 00:01:51,000 --> 00:01:54,680 Speaker 1: Else, they really have. They've been documenting this explosion of 39 00:01:54,760 --> 00:01:57,799 humanoid robotics, capturing all the little details that make you 40 00:01:57,840 --> 00:02:00,439 just stop and go wait. Is this actually real? 41 00:02:00,719 --> 00:02:02,640 Speaker 2: And that is the question We're gonna be asking a 42 00:02:02,680 --> 00:02:05,640 lot today. Is this real? And if it is, what 43 00:02:05,680 --> 00:02:07,760 does it mean for the person listening to this right now? 44 00:02:08,360 --> 00:02:12,319 Because it's one thing to watch a roll doun on YouTube, right, 45 00:02:12,520 --> 00:02:16,280 it's another thing entirely to realize it's training to be 46 00:02:16,360 --> 00:02:18,080 your coworker or your opponent. 47 00:02:18,479 --> 00:02:20,639 Speaker 1: Okay, let's not waste any time. We have to start 48 00:02:20,639 --> 00:02:22,840 with the violence. I mean, we just have to. I 49 00:02:22,919 --> 00:02:26,560 need to talk about this combat league Engine AI, Engine AI. 50 00:02:27,240 --> 00:02:31,759 They have audienced the world's first commercial humanoid free combat league. 51 00:02:32,120 --> 00:02:36,479 Just saying those words together feels wild. Commercial humanoid combat. 52 00:02:36,599 --> 00:02:39,080 Speaker 2: It is a significant pivot for the industry. Usually when 53 00:02:39,080 --> 00:02:41,520 we talk about robots, we picture them, you know, carefully 54 00:02:41,639 --> 00:02:43,240 navigating obstacle courses in a. 55 00:02:43,280 --> 00:02:45,759 Speaker 1: Lab, or folding laundry really really. 56 00:02:45,520 --> 00:02:48,159 Speaker 2: Slowly, or slowly folding a shirt while a grad student 57 00:02:48,199 --> 00:02:51,319 holds their breath praying it doesn't drop it. Engine AI 58 00:02:51,520 --> 00:02:54,159 has decided to go the route of the Roman Colisseum. 59 00:02:54,240 --> 00:02:56,840 They launched this initiative just recently, and the premise is 60 00:02:56,840 --> 00:03:00,560 effectively the UFC, but you replace the flesh and with 61 00:03:00,719 --> 00:03:01,719 metal and code. 62 00:03:01,960 --> 00:03:04,599 Speaker 1: And the hardware they're using is called the T eight hundred. 63 00:03:04,759 --> 00:03:07,879 And I assume for anyone listening who loves eighty cinema, 64 00:03:07,960 --> 00:03:09,719 that name is absolutely intentional. 65 00:03:09,800 --> 00:03:13,520 Speaker 2: Oh completely, They are not hiding the terminator influence. They 66 00:03:13,520 --> 00:03:14,599 are leaning into it hard. 67 00:03:14,639 --> 00:03:15,439 Speaker 1: I'm embracing it. 68 00:03:15,680 --> 00:03:18,199 Speaker 2: The whole philosophy behind the T eight hundred is to 69 00:03:18,479 --> 00:03:23,080 mimic the structural proportions of an adult human perfectly that 70 00:03:23,159 --> 00:03:26,199 one hundred and sixty five pounds sweet spot, but build 71 00:03:26,240 --> 00:03:30,599 it with materials that allow for well superhuman resilience. 72 00:03:31,000 --> 00:03:33,360 Speaker 1: So walk us through the league itself, because when I 73 00:03:33,360 --> 00:03:35,400 first read about this, I just assumed it was like 74 00:03:35,719 --> 00:03:38,479 engine AI fighting their own robots against each other, you know, 75 00:03:38,520 --> 00:03:41,000 a tech demo where the outcome is kind of rigged. 76 00:03:41,159 --> 00:03:43,680 Speaker 2: No, that would be boring, that's just marketing. This is 77 00:03:43,719 --> 00:03:47,520 a sport, a real competition, real competition. They have established 78 00:03:47,520 --> 00:03:49,719 a tiered tournament that's going to run all the way 79 00:03:49,719 --> 00:03:53,439 through December. There are sixteen teams involved. And this is 80 00:03:53,439 --> 00:03:57,120 the interesting part. On one side, you have these elite 81 00:03:57,199 --> 00:04:00,759 university laboratories, the academics that people writing the recent search papers. 82 00:04:00,919 --> 00:04:05,400 On the other side, you have hardcore robotic startups, the industry, 83 00:04:05,439 --> 00:04:09,560 the people building the actual products. It's academia versus industry, ye, 84 00:04:09,759 --> 00:04:10,719 battling it out in the ring. 85 00:04:10,879 --> 00:04:13,400 Speaker 1: That's a great setup. But here's the part that really 86 00:04:13,439 --> 00:04:16,759 made my jaw drop. Usually a robot like this costs 87 00:04:17,000 --> 00:04:19,480 what I mean, it's got to be high six figures. 88 00:04:19,199 --> 00:04:22,720 Speaker 2: Right, oh easily for a prototype of this caliber. With 89 00:04:22,839 --> 00:04:26,079 these kinds of custom actuators and sensors, you are looking 90 00:04:26,079 --> 00:04:29,800 at hundreds of thousands of dollars, maybe even touching a million, 91 00:04:29,839 --> 00:04:31,480 depending on the supply chain in R and D. 92 00:04:31,839 --> 00:04:35,040 Speaker 1: Right. But engine AI is giving the robots to these 93 00:04:35,040 --> 00:04:38,120 teams for free. They are They are literally handing over 94 00:04:38,199 --> 00:04:40,839 millions of dollars in hardware and just saying, here you go, 95 00:04:40,839 --> 00:04:42,560 go program this thing to fight. 96 00:04:42,879 --> 00:04:46,000 Speaker 2: It seems counterintuitive from a business standpoint, doesn't it. Why 97 00:04:46,040 --> 00:04:48,240 give away your most advanced product just to watch it 98 00:04:48,240 --> 00:04:49,439 get smashed to pieces? 99 00:04:49,560 --> 00:04:51,800 Speaker 1: Exactly? I get nervous lending someone in my car. These 100 00:04:51,800 --> 00:04:53,759 guys are handing over a T eight hundred and saying, 101 00:04:54,040 --> 00:04:57,439 hope you've coded a good roundhouse kick. Is this just 102 00:04:57,519 --> 00:04:59,639 a massive, expensive PR stunt. 103 00:05:00,040 --> 00:05:04,560 Speaker 2: It's actually a very very calculated strategy disguised as a spectacle. 104 00:05:04,759 --> 00:05:08,319 It all comes down to one word. Data. Engine AI 105 00:05:08,560 --> 00:05:11,519 isn't just building a sport. They are building a stress 106 00:05:11,639 --> 00:05:14,759 test that they could never possibly replicate in a lab. 107 00:05:15,079 --> 00:05:17,399 What do you mean, well, think about it. A factory 108 00:05:17,399 --> 00:05:21,040 floor is dangerous, sure, a disaster zone is dangerous. But 109 00:05:21,199 --> 00:05:25,279 nothing is as unpredictable as chaotic as a combat scenario 110 00:05:25,560 --> 00:05:28,439 against another intelligent agent that is actively trying to knock 111 00:05:28,480 --> 00:05:28,839 you down. 112 00:05:29,160 --> 00:05:33,000 Speaker 1: So by watching these robots fight, they're crowdsourcing the development 113 00:05:33,120 --> 00:05:35,040 of more survival instincts. 114 00:05:35,120 --> 00:05:38,800 Speaker 2: Precisely, think about the data points you're collecting. If a 115 00:05:38,879 --> 00:05:41,480 robot can survive a full force roundhouse kick to the 116 00:05:41,600 --> 00:05:43,439 chassis and stay standing. 117 00:05:43,160 --> 00:05:45,720 Speaker 1: Then it can probably survive a forklift bumping into it 118 00:05:45,720 --> 00:05:46,480 at the warehouse. 119 00:05:46,519 --> 00:05:49,160 Speaker 2: That's exactly it. If it can take a fall, get 120 00:05:49,199 --> 00:05:51,959 back up in under ten seconds and re engage, then 121 00:05:51,959 --> 00:05:54,600 it could probably survive a collapsing wall in a search 122 00:05:54,639 --> 00:05:59,160 and rescue mission. They're accelerating evolution by introducing predation. They're 123 00:05:59,199 --> 00:06:02,759 finding every week, every software bug, every structural flaw by 124 00:06:02,800 --> 00:06:05,120 letting other smart people try to break their stuff. 125 00:06:05,439 --> 00:06:10,120 Speaker 1: That is actually brilliant, terrifying, but brilliant. And I saw 126 00:06:10,160 --> 00:06:12,000 there's real money on the line too. This isn't just 127 00:06:12,040 --> 00:06:13,199 for bragging rights. 128 00:06:13,079 --> 00:06:15,560 Speaker 2: No, not at all. We're talking about a prize pot 129 00:06:15,759 --> 00:06:19,560 of one point four to four million dollars and a belt. 130 00:06:19,839 --> 00:06:21,160 Speaker 1: Oh tell me about the belt. 131 00:06:21,240 --> 00:06:25,120 Speaker 2: It's a championship belt made of ten kilograms of solid gold. 132 00:06:25,360 --> 00:06:29,040 Speaker 1: Ten kilos that's over twenty pounds of Gold's that's just 133 00:06:29,120 --> 00:06:29,920 heavy valued. 134 00:06:29,959 --> 00:06:33,480 Speaker 2: It roughly ten million r MB. It's pure spectacle, designed 135 00:06:33,480 --> 00:06:35,800 to draw eyes and make this feel like a legitimate 136 00:06:35,879 --> 00:06:36,759 high stake sport. 137 00:06:37,000 --> 00:06:39,439 Speaker 1: Wow. Okay, So what about the rules. How does a 138 00:06:39,519 --> 00:06:40,519 robot fight work? 139 00:06:40,639 --> 00:06:43,519 Speaker 2: The rules are what make it interesting. From a technical standpoint, 140 00:06:43,720 --> 00:06:46,040 it's a best of three format like a tennis match. 141 00:06:46,560 --> 00:06:48,959 If a robot gets knocked down, it has ten seconds 142 00:06:49,000 --> 00:06:51,199 to stand back up on its own. If it can't, 143 00:06:51,279 --> 00:06:54,360 it's a knockout, just like boxing exactly. And here's the 144 00:06:54,399 --> 00:06:57,639 real kicker for the engineers. If the human operators have 145 00:06:57,759 --> 00:07:00,480 to physically step in to assist or reset the bot 146 00:07:00,519 --> 00:07:02,920 more than twice, that's a TKO. 147 00:07:02,639 --> 00:07:03,680 Speaker 1: A technical knockout. 148 00:07:03,800 --> 00:07:07,240 Speaker 2: Yep. So the goal is total autonomy. You can't have 149 00:07:07,279 --> 00:07:09,560 a human running out there every five seconds to pick 150 00:07:09,560 --> 00:07:11,959 it up like a toddler who just fell over. It 151 00:07:12,000 --> 00:07:14,519 has to be self sufficient in a hostile environment. 152 00:07:15,040 --> 00:07:18,519 Speaker 1: Speaking of autonomy and stability, we have to talk about 153 00:07:18,639 --> 00:07:22,920 the CEO Shaotanyang. We do because when the first footage 154 00:07:23,000 --> 00:07:26,040 of the T eight hundred came out. The internet did 155 00:07:26,040 --> 00:07:31,519 what the Internet does. This whole thing immediately cgi human 156 00:07:31,600 --> 00:07:32,199 in a suit. 157 00:07:32,639 --> 00:07:35,639 Speaker 2: It happens every time a major leap in robotics occurs. 158 00:07:36,079 --> 00:07:39,720 We call it the Uncanny Valley of movement. Skepticism is natural. 159 00:07:39,800 --> 00:07:42,079 You know, the movements look too fluid, the spinning kicks 160 00:07:42,079 --> 00:07:44,319 were too perfect. People just couldn't believe it was real. 161 00:07:44,800 --> 00:07:48,319 Speaker 1: So Chao Tanyang decides to debunk this, but he doesn't 162 00:07:48,319 --> 00:07:50,319 issue a press release, He doesn't do a boring Q 163 00:07:50,439 --> 00:07:52,680 and A with a PowerPoint. He gets in the ring. 164 00:07:52,879 --> 00:07:55,000 Speaker 2: He did, and this is a testament to either his 165 00:07:55,079 --> 00:07:59,319 incredible bravery or his absolute unshakable faith in his engineering team, 166 00:07:59,439 --> 00:08:01,839 or both or both. Yeah, he suited up in this 167 00:08:01,920 --> 00:08:05,240 heavy protective gear, I mean serious riot gear level padding 168 00:08:05,680 --> 00:08:07,160 and stood in front of the T eight hundred. 169 00:08:07,360 --> 00:08:11,040 Speaker 1: I saw the footage of this. It's haunting. The robot 170 00:08:11,079 --> 00:08:13,600 isn't moving like a machine, It's moving like a predator. 171 00:08:14,040 --> 00:08:17,040 It stabilizes its stance. You can almost see it calculating 172 00:08:17,079 --> 00:08:20,639 the distance and then bam, a high torque sidekick right 173 00:08:20,680 --> 00:08:21,360 to his chest. 174 00:08:21,439 --> 00:08:24,000 Speaker 2: And let's be really clear about the physics here. This 175 00:08:24,160 --> 00:08:26,480 was not a tap, this wasn't a love pad. The 176 00:08:26,560 --> 00:08:30,720 impact literally lifted the CEO off the ground. 177 00:08:30,839 --> 00:08:34,240 Speaker 1: He was airborne for a split second. Both his feet 178 00:08:34,240 --> 00:08:35,080 are off the mat. 179 00:08:35,200 --> 00:08:37,200 Speaker 2: He dropped straight onto the mat. There were no wires, 180 00:08:37,240 --> 00:08:40,559 no hidden rigs, no special effects, just pure kinetic energy 181 00:08:40,559 --> 00:08:42,879 transfer from a metal leg to a human. 182 00:08:42,679 --> 00:08:45,799 Speaker 1: Torso and the robot did it fall over because Newton's 183 00:08:45,799 --> 00:08:48,320 third law says for every action there is an equal 184 00:08:48,320 --> 00:08:50,559 and opposite reaction. If it kicks him that hard, it 185 00:08:50,559 --> 00:08:51,559 should fly backward. 186 00:08:52,159 --> 00:08:55,080 Speaker 2: That is the key engineering detail. That's the magic. After 187 00:08:55,120 --> 00:08:57,840 delivering that massive amount of force, that T eight hundred 188 00:08:57,960 --> 00:09:01,360 didn't wobble, It didn't stumble backward, didn't even flinch nothing. 189 00:09:01,480 --> 00:09:05,120 Its gyroscopic sensors and its control loops reset its balance 190 00:09:05,120 --> 00:09:08,080 in milliseconds. It struck, and it returned to a ready 191 00:09:08,120 --> 00:09:10,120 stance immediately. It was as if nothing had happened. 192 00:09:10,320 --> 00:09:13,120 Speaker 1: That is the part that scares me. The CEO's reaction 193 00:09:13,200 --> 00:09:15,720 was pretty telling too. He gets up and through his 194 00:09:15,720 --> 00:09:17,519 helmet you can hear him say too violent. 195 00:09:17,840 --> 00:09:20,720 Speaker 2: He literally joked that he couldn't guarantee he'd stay alive 196 00:09:20,759 --> 00:09:21,799 for the next test. 197 00:09:21,840 --> 00:09:23,600 Speaker 1: Which I mean is a pretty good way to prove 198 00:09:23,639 --> 00:09:24,080 your point. 199 00:09:24,240 --> 00:09:27,159 Speaker 2: It was an incredible stress demonstration. It proved the robot 200 00:09:27,200 --> 00:09:30,120 is built for massive impact, but it also proved that 201 00:09:30,159 --> 00:09:33,879 the T eight hundred is fundamentally a weapon. In that context, 202 00:09:33,919 --> 00:09:36,080 it's not a helper at that moment. It's a striker. 203 00:09:36,200 --> 00:09:38,039 Speaker 1: Okay, let's get under the hood of this thing. What 204 00:09:38,240 --> 00:09:40,679 makes it capable of kicking a grown man into the 205 00:09:40,720 --> 00:09:42,240 air and not even wobbling. 206 00:09:42,440 --> 00:09:45,480 Speaker 2: It really comes down to torque and power management. So 207 00:09:45,720 --> 00:09:47,519 each joint motor in the T. 208 00:09:47,559 --> 00:09:51,000 Speaker 1: Eight hundred, like the knee, the hip, the shoulder. 209 00:09:50,759 --> 00:09:53,720 Speaker 2: Exactly, each of those joints delivers up to four hundred 210 00:09:53,720 --> 00:09:55,720 and fifty newton meters of torque. 211 00:09:56,000 --> 00:09:59,200 Speaker 1: Okay, pause there, for those of us who aren't mechanical engineers. 212 00:09:59,600 --> 00:10:02,279 Put four hundred and fifty newton meters in perspective, is 213 00:10:02,320 --> 00:10:02,919 that a lot? 214 00:10:03,039 --> 00:10:05,080 Speaker 2: It is an immense amount for a joint the size 215 00:10:05,080 --> 00:10:07,320 of a human knee. To give you a comparison, a 216 00:10:07,399 --> 00:10:10,519 standard family set an engine, a four cylinder engine might 217 00:10:10,559 --> 00:10:13,159 produce anywhere from two hundred to four hundred newton meters 218 00:10:13,159 --> 00:10:15,000 of torque to move the entire car. 219 00:10:14,960 --> 00:10:15,879 Speaker 1: The whole two ton car. 220 00:10:16,080 --> 00:10:18,759 Speaker 2: Yes, yeah, we are talking about that kind of force 221 00:10:18,799 --> 00:10:22,399 potential in a single joint. It is enough to generate 222 00:10:22,480 --> 00:10:27,080 bone shattering momentum. The peak power output is instantaneous fourteen 223 00:10:27,120 --> 00:10:30,600 thousand watts. It's just an explosion of energy on demand. 224 00:10:30,679 --> 00:10:32,799 Speaker 1: But usually when you run that much power through a 225 00:10:32,840 --> 00:10:36,679 motor that's small, it gets hot like melt, the plastic hot. 226 00:10:36,759 --> 00:10:40,679 Speaker 2: Correct thermal throttling is the enemy of high performance robotics. 227 00:10:41,440 --> 00:10:44,279 Usually a robot can do a cool trick once or twice, 228 00:10:44,600 --> 00:10:46,399 and then it has to sit and cool down for 229 00:10:46,480 --> 00:10:48,120 ten minutes so it doesn't fry its own. 230 00:10:47,919 --> 00:10:49,320 Speaker 1: Circuits, right, it overheats. 231 00:10:49,440 --> 00:10:52,039 Speaker 2: The T eight hundred features a world first active cooling 232 00:10:52,080 --> 00:10:53,480 system built directly into. 233 00:10:53,360 --> 00:10:56,039 Speaker 1: The leg joints, so it's like a radiator for its knees. 234 00:10:56,320 --> 00:10:59,840 Speaker 2: That is a perfect analogy. It circulates coolant to keep 235 00:10:59,879 --> 00:11:02,799 the motors at optimal temperatures even under extreme load. 236 00:11:02,840 --> 00:11:05,320 Speaker 1: And what does that mean In practical terms, it means. 237 00:11:05,080 --> 00:11:07,879 Speaker 2: It can fight, or work or run for up to 238 00:11:07,919 --> 00:11:11,080 four hours without thermal throttling. It's the difference between a 239 00:11:11,120 --> 00:11:14,080 sprinter who can only run one hundred meters and a 240 00:11:14,120 --> 00:11:16,960 marathon runner who can sprint the entire way it has 241 00:11:17,039 --> 00:11:18,120 endurance that is. 242 00:11:18,120 --> 00:11:20,480 Speaker 1: A total game changer. And the brain running all this. 243 00:11:20,600 --> 00:11:21,279 What's it using. 244 00:11:21,600 --> 00:11:24,879 Speaker 2: It's using the Nvidia jets and Thord chip. This is 245 00:11:24,919 --> 00:11:31,200 an absolute powerhouse of AI computing. It delivers two thousand tops. 246 00:11:30,399 --> 00:11:32,799 Speaker 1: Sorry tops, what's that tops? 247 00:11:33,120 --> 00:11:36,200 Speaker 2: It stands for trillions of operations per second? 248 00:11:36,360 --> 00:11:39,440 Speaker 1: Two two thousand trillion operations per second. 249 00:11:39,519 --> 00:11:42,879 Speaker 2: Yes, it's a staggering amount of compute power. It allows 250 00:11:42,919 --> 00:11:46,360 the robot to process its entire environment through three hundred 251 00:11:46,360 --> 00:11:50,559 and sixty degree lightar and stereodepth cameras in milliseconds. It 252 00:11:50,639 --> 00:11:52,919 isn't just seeing pixels on a screen. 253 00:11:53,120 --> 00:11:54,720 Speaker 1: It's perceiving the world. 254 00:11:54,960 --> 00:11:59,399 Speaker 2: It is perceiving geometry. It understands depth, velocity, and trajectory 255 00:11:59,440 --> 00:12:02,000 in real time. That's how it pulls off those cappaera 256 00:12:02,120 --> 00:12:05,240 style transitions and spinning kicks. It knows exactly where its 257 00:12:05,240 --> 00:12:06,960 limbs are and where its opponent is in three D 258 00:12:07,080 --> 00:12:08,039 space at all times. 259 00:12:08,120 --> 00:12:09,919 Speaker 1: I noticed that in the video it doesn't do the 260 00:12:10,000 --> 00:12:12,600 robot shuffle, you know that bent knee crouching walk that 261 00:12:12,679 --> 00:12:13,840 every robot seems to do. 262 00:12:13,919 --> 00:12:14,919 Speaker 2: The Groucho Marx walk. 263 00:12:15,000 --> 00:12:16,960 Speaker 1: Yeah, yes, why do they do that? 264 00:12:16,960 --> 00:12:19,399 Speaker 2: That's done for stability. It keeps the center of gravity 265 00:12:19,440 --> 00:12:22,480 low so the robot doesn't tip over easily. It's a safer, 266 00:12:22,600 --> 00:12:25,759 but much less efficient and less human like way to walk. 267 00:12:25,840 --> 00:12:27,279 Speaker 1: But the T eight hundred doesn't. 268 00:12:27,480 --> 00:12:30,399 Speaker 2: No, the T eight hundred uses straight leg walking. It 269 00:12:30,440 --> 00:12:32,799 looks much more human, much more natural, but it is 270 00:12:32,879 --> 00:12:35,639 infinitely harder to control because your center of gravity is 271 00:12:35,720 --> 00:12:38,919 higher and constantly shifting. The fact that they have nailed 272 00:12:38,919 --> 00:12:42,480 straight leg walking while also delivering combat level kicks is 273 00:12:42,519 --> 00:12:45,240 a massive engineering flex. It's showing off. 274 00:12:45,240 --> 00:12:47,360 Speaker 1: It really is, But you touched on something earlier the 275 00:12:47,480 --> 00:12:50,960 why behind all this engine AI claims this combat first 276 00:12:50,960 --> 00:12:53,720 approach speeds up their R and D cycle by thirty percent. 277 00:12:54,279 --> 00:12:56,440 But are we kind of ignoring the elephant in the 278 00:12:56,480 --> 00:12:59,000 room here? We are training these systems to be lethal 279 00:12:59,000 --> 00:12:59,799 fighters first. 280 00:13:00,080 --> 00:13:02,960 Speaker 2: It is a valid and very important concern. If the 281 00:13:03,000 --> 00:13:06,480 foundational data set for this robot's movement, for its instincts 282 00:13:06,679 --> 00:13:09,600 comes from violence and combat, what does that mean for 283 00:13:09,639 --> 00:13:10,960 its future role in society? 284 00:13:11,120 --> 00:13:13,840 Speaker 1: Right? We aren't training them to help gramdmas cross the street. 285 00:13:13,840 --> 00:13:16,399 We're training them to sweep the leg exactly. 286 00:13:17,240 --> 00:13:20,120 Speaker 2: Is the T eight hundred a breakthrough in robust engineering, 287 00:13:20,559 --> 00:13:22,600 or are we just laying the foundation for a very 288 00:13:22,600 --> 00:13:27,480 specific type of enforcement. Droid history suggests that technology developed 289 00:13:27,480 --> 00:13:31,480 for combat eventually trickles down to civilian use, you know, GPS, 290 00:13:31,519 --> 00:13:34,879 the Internet, jet engines, but the DNA of that technology, 291 00:13:34,919 --> 00:13:38,039 its origin often remains. It's something we need to watch 292 00:13:38,159 --> 00:13:39,000 very very closely. 293 00:13:39,039 --> 00:13:43,080 Speaker 1: Well, speaking of DNA and evolution, let's pivot from the 294 00:13:43,200 --> 00:13:45,480 violence of the ring to the agility of the street. 295 00:13:45,559 --> 00:13:47,240 Speaker 2: A complete tonal shift. 296 00:13:47,039 --> 00:13:49,519 Speaker 1: Because while the T eight hundred is out there breaking bones, 297 00:13:49,879 --> 00:13:52,919 or at least simulated bones, another robot is out there 298 00:13:52,960 --> 00:13:54,039 shredding the asphalt. 299 00:13:54,240 --> 00:13:57,000 Speaker 2: Yes, the Unitary g one. We are moving from the 300 00:13:57,039 --> 00:14:00,240 heavy industrial violence of the T eight hundred to extremem 301 00:14:00,240 --> 00:14:02,879 agility and frankly, a sense of playfulness. 302 00:14:02,960 --> 00:14:05,600 Speaker 1: Unitru's g one robot just learned to skateboard. And I 303 00:14:05,639 --> 00:14:07,759 don't mean it stood on a moving board someone pushed. 304 00:14:07,879 --> 00:14:09,440 I mean it learned to skate right. 305 00:14:09,679 --> 00:14:11,879 Speaker 2: And crucially, this was not a staged lab trick with 306 00:14:11,879 --> 00:14:13,799 a safety tedt or holding it up in a perfectly 307 00:14:13,799 --> 00:14:16,399 smooth floor. This was out in the open world, on 308 00:14:16,480 --> 00:14:20,360 real asphalt, dealing with friction and cracks and uneven surfaces. 309 00:14:20,559 --> 00:14:23,639 Speaker 1: This was a collaboration, right. It wasn't just unitry working 310 00:14:23,679 --> 00:14:24,360 in a vacuum. 311 00:14:24,480 --> 00:14:27,799 Speaker 2: No, it was a massive joint effort. Unitary partnered with 312 00:14:27,919 --> 00:14:32,639 China Telecom's Tealai Institute and also Shanghai jou Tan University. 313 00:14:33,039 --> 00:14:36,840 They developed a whole new control framework for this called Husky. 314 00:14:37,039 --> 00:14:39,080 Speaker 1: Husky I like that sounds sturdy. 315 00:14:39,600 --> 00:14:43,039 Speaker 2: It stands for a physics aware whole body control system, 316 00:14:43,399 --> 00:14:46,039 and that physics wear part is the entire secret. 317 00:14:46,039 --> 00:14:47,919 Speaker 1: Okay, break that down for us. What makes it different 318 00:14:47,919 --> 00:14:50,519 from just programming an animation that says move left leg 319 00:14:50,600 --> 00:14:51,759 now than move right arm. 320 00:14:51,960 --> 00:14:54,399 Speaker 2: So when you program an animation, the robot is just 321 00:14:54,519 --> 00:14:57,120 hitting its marks. It's like following a dance chart. It 322 00:14:57,159 --> 00:14:59,639 doesn't know why it's moving that way. It's just executing 323 00:14:59,639 --> 00:15:02,759 a seat quince. The Husky framework allows the g one 324 00:15:03,159 --> 00:15:06,639 to actually reason about physics in real time. 325 00:15:06,759 --> 00:15:08,080 Speaker 1: It understands the world around it. 326 00:15:08,200 --> 00:15:10,360 Speaker 2: It understands that if it shifts its weight to the left, 327 00:15:10,440 --> 00:15:13,840 the board will tilt, the wheels will turn, and centrifugal 328 00:15:13,919 --> 00:15:15,200 force will act on its body. 329 00:15:15,399 --> 00:15:18,840 Speaker 1: It's not just moving, it's balancing, so it's feeling the ride. 330 00:15:18,639 --> 00:15:21,879 Speaker 2: In a mathematical sense. Yes, it isn't just replaying a 331 00:15:21,919 --> 00:15:24,919 script from memory. It is constantly reacting to the board 332 00:15:24,960 --> 00:15:28,120 under its feet. If the board hits a pebble and wobbles, 333 00:15:28,440 --> 00:15:33,559 the robot instinctively or rather computationally, adjusts its center of 334 00:15:33,639 --> 00:15:36,039 mass in milliseconds to compensate. 335 00:15:36,399 --> 00:15:38,759 Speaker 1: That is just wild, and the way they taught it 336 00:15:38,840 --> 00:15:41,279 is fascinating too. This whole sim to real thing. 337 00:15:41,600 --> 00:15:44,279 Speaker 2: This is the secret sauce of modern robotics. This is 338 00:15:44,279 --> 00:15:47,159 how they're learning so fast. You don't teach a robot 339 00:15:47,200 --> 00:15:49,320 to skateboard by pushing it down a hill one thousand 340 00:15:49,360 --> 00:15:53,159 dimes and breaking a thousand very expensive robots. You build 341 00:15:53,240 --> 00:15:57,960 a perfect physics simulation, a video game essentially, that perfectly 342 00:15:57,960 --> 00:16:01,639 mimics the real world gravity, friction, win, resistance. 343 00:16:01,200 --> 00:16:03,120 Speaker 1: All of it, and you just let the digital version 344 00:16:03,159 --> 00:16:06,440 of the robot fail millions of times inside the computer exactly. 345 00:16:06,440 --> 00:16:09,360 Speaker 2: They ran thousands of parallel training environments at the same time. 346 00:16:09,720 --> 00:16:13,000 They compress millions and millions of motion trials into just 347 00:16:13,080 --> 00:16:14,440 twenty hours of compute time. 348 00:16:14,519 --> 00:16:17,480 Speaker 1: Twenty hours, that's it. A human takes years to get 349 00:16:17,559 --> 00:16:17,919 that good. 350 00:16:18,240 --> 00:16:21,399 Speaker 2: That's the power of reinforcement learning in less than a 351 00:16:21,480 --> 00:16:24,759 day of processing, The AI learned the equivalent of years 352 00:16:24,799 --> 00:16:28,600 of human practice through trial and error at an accelerated pace. 353 00:16:28,840 --> 00:16:32,960 Speaker 1: That speed is well, it's terrifyingly fast. I mean, imagine 354 00:16:33,000 --> 00:16:34,639 if I could learn to play the piano in twenty 355 00:16:34,679 --> 00:16:37,080 hours just by thinking about it really really hard. 356 00:16:37,159 --> 00:16:39,960 Speaker 2: It's the matrix concept, isn't it. I know, kung fu. 357 00:16:40,159 --> 00:16:42,720 Once the algorithm solves the physics of the task and 358 00:16:42,759 --> 00:16:45,960 the simulation, they just transfer that brain directly into the 359 00:16:45,960 --> 00:16:49,960 physical g one robot. And because the simulation was so accurate, 360 00:16:50,159 --> 00:16:52,639 the robot steps onto the real skateboard and just goes. 361 00:16:52,840 --> 00:16:55,039 Speaker 1: And what can it actually do? It's not just like 362 00:16:55,440 --> 00:16:57,120 rolling forward in a straight line. 363 00:16:57,200 --> 00:16:59,840 Speaker 2: Oh no, it's actively pushing off with one leg to 364 00:16:59,840 --> 00:17:03,279 get speed, which requires balancing on one foot on a 365 00:17:03,279 --> 00:17:06,599 moving object. By the way, it's steering by leaning its body. 366 00:17:07,079 --> 00:17:09,240 It can mount and dismount the board while the board 367 00:17:09,279 --> 00:17:12,720 is still moving, and crucially, it can recover its balance 368 00:17:12,759 --> 00:17:13,799 when someone nudges it. 369 00:17:14,200 --> 00:17:17,039 Speaker 1: That nudge recovery is huge. That means it's robust, it's 370 00:17:17,079 --> 00:17:17,720 not fragile. 371 00:17:17,960 --> 00:17:20,440 Speaker 2: It implies that once a robot learns the physics of 372 00:17:20,440 --> 00:17:23,839 a task, any task it can master it faster than 373 00:17:23,839 --> 00:17:28,119 a human ever could. Skateboarding is just the cool demo. Yeah, 374 00:17:28,359 --> 00:17:31,359 now apply that same learning process to walking on the 375 00:17:31,359 --> 00:17:35,079 deck of a tossing ship or navigating a cluttered construction site. 376 00:17:35,279 --> 00:17:37,079 The implications are endless. 377 00:17:37,359 --> 00:17:39,559 Speaker 1: So we have the fighter and we have the skater. 378 00:17:39,720 --> 00:17:41,519 But let's talk about the one that's actually going to 379 00:17:41,599 --> 00:17:43,559 pay the bills, the worker. 380 00:17:43,480 --> 00:17:45,480 Speaker 2: That brings us to the titan of the industry. 381 00:17:46,079 --> 00:17:49,559 Speaker 1: Boston Dynamics, the OG, the one everyone knows, the. 382 00:17:49,519 --> 00:17:53,559 Speaker 2: OG, and they have completely utterly leveled up with the 383 00:17:53,599 --> 00:17:54,160 new outlics. 384 00:17:54,279 --> 00:17:56,440 Speaker 1: Now, there was a bit of drama here, right because 385 00:17:56,680 --> 00:17:58,759 back in early January there was a leak. 386 00:17:59,079 --> 00:18:02,039 Speaker 2: There was some footage surface of the new Atlas looking 387 00:18:02,640 --> 00:18:05,759 well shaky, Yeah, not great. It looked unstable. There was 388 00:18:05,880 --> 00:18:09,279 visible damage to the palm mechanisms, the hands. The Internet 389 00:18:09,279 --> 00:18:11,880 did what the Internet does and immediately declared that Boston 390 00:18:11,960 --> 00:18:14,920 Dynamics was years away from a production model. I said, look, 391 00:18:15,240 --> 00:18:16,920 it's broken. They've lost their touch. 392 00:18:17,119 --> 00:18:21,359 Speaker 1: And then one month later CES happens, the big tech show, 393 00:18:21,680 --> 00:18:22,279 and they just. 394 00:18:22,359 --> 00:18:26,079 Speaker 2: Dropped the mic. They revealed the production ready electric atlics 395 00:18:27,079 --> 00:18:30,920 and it wasn't shaky. It was flawless, and it's big. 396 00:18:31,000 --> 00:18:33,519 Speaker 1: I was reading the specs. Seven and a half feet tall. 397 00:18:33,640 --> 00:18:34,680 Speaker 2: That is massive? 398 00:18:34,880 --> 00:18:39,920 Speaker 1: Why so big? That seems impractical for a human world. 399 00:18:40,160 --> 00:18:41,720 It won't fit through a normal door. 400 00:18:41,880 --> 00:18:44,119 Speaker 2: It's not designed for our living rooms. It's designed for 401 00:18:44,200 --> 00:18:48,200 warehouses and factories. It's about leverage and reach. If you're 402 00:18:48,240 --> 00:18:50,599 building a factory robot, you wanted to be able to 403 00:18:50,640 --> 00:18:55,119 reach high shelves, handle large, heavy components, and exert serious force. 404 00:18:55,720 --> 00:18:58,480 At seven point five feet, it towers over human workers, 405 00:18:58,720 --> 00:19:01,440 but it fits into the industrial scale perfectly. It can 406 00:19:01,480 --> 00:19:03,559 reach things you'd need a forklift or a ladder for. 407 00:19:03,839 --> 00:19:07,119 Speaker 1: And it's electric now, not hydraulic, because the old Atlas 408 00:19:07,160 --> 00:19:09,799 was famous for being really loud, all those hydraulic fluids 409 00:19:09,799 --> 00:19:10,279 and pumps. 410 00:19:10,359 --> 00:19:13,440 Speaker 2: Yes, the switch to electric actuators is a huge deal. 411 00:19:13,759 --> 00:19:16,319 That means it's much quieter, it's cleaner, and it's far 412 00:19:16,400 --> 00:19:18,519 easier to maintain. You don't have to worry about leaking 413 00:19:18,599 --> 00:19:20,920 hydraulic oil all over the factory floor. 414 00:19:21,000 --> 00:19:22,640 Speaker 1: And the strength, what can it do? 415 00:19:23,039 --> 00:19:26,319 Speaker 2: It can handle one hundred and ten pound loads, which 416 00:19:26,359 --> 00:19:28,640 is significant if you're a human and you have to 417 00:19:28,680 --> 00:19:30,599 lift one hundred and ten pounds over and over again 418 00:19:30,640 --> 00:19:33,680 all day. You're going to blow out your back. Atlas 419 00:19:33,720 --> 00:19:35,640 can do it twenty four to seven without a coffee break. 420 00:19:35,680 --> 00:19:38,200 Speaker 1: And it can operate in extreme temperatures too. I saw 421 00:19:38,240 --> 00:19:40,920 from minus four degrees fahrenheit up to one hundred and 422 00:19:40,920 --> 00:19:42,200 four degrees so it can work in. 423 00:19:42,160 --> 00:19:45,279 Speaker 2: A giant freezer warehouse or a sweltering foundry without a 424 00:19:45,319 --> 00:19:49,480 single complaint. But the physical specs, as impressive as they are, 425 00:19:49,599 --> 00:19:51,119 aren't even the most amazing part. 426 00:19:51,200 --> 00:19:51,759 Speaker 1: It's the brain. 427 00:19:51,880 --> 00:19:54,359 Speaker 2: It's the brain. This is the partnership with Google Deep 428 00:19:54,440 --> 00:19:58,920 Mind using their Gemini AI. Boston Dynamics has always had 429 00:19:58,960 --> 00:20:02,200 the best bodies in the businesiness, the best hardware, but 430 00:20:02,279 --> 00:20:06,839 their brains were often hard coded, very specific, rigid instructions 431 00:20:06,880 --> 00:20:11,319 for specific tasks. Now they've integrated Gemini, and this means 432 00:20:11,319 --> 00:20:14,200 Atlas isn't just programmed to do step A then step B. 433 00:20:14,839 --> 00:20:15,920 It is problem solving. 434 00:20:16,119 --> 00:20:18,440 Speaker 1: So if a box is slightly out of place, it 435 00:20:18,480 --> 00:20:20,960 doesn't just freeze and throw an error message exactly. 436 00:20:21,079 --> 00:20:23,799 Speaker 2: It observes the situation, It reasons about what it needs 437 00:20:23,799 --> 00:20:27,279 to do, and it adjusts its plan. They claim it 438 00:20:27,279 --> 00:20:30,480 can master entirely new factory tasks. In under twenty four hours, 439 00:20:30,559 --> 00:20:33,000 just by observing a human do it. Wait what, You 440 00:20:33,039 --> 00:20:35,319 don't have to write thousands of lines of new code. 441 00:20:35,359 --> 00:20:37,400 You just show it what to do and the learns. 442 00:20:37,519 --> 00:20:41,480 Speaker 1: That is the holy grail of automation. Watch me then 443 00:20:41,599 --> 00:20:43,440 do it forever without getting tired. 444 00:20:43,599 --> 00:20:46,720 Speaker 2: It is, and the scale of production here is serious. 445 00:20:47,000 --> 00:20:50,039 This isn't lab toy anymore. Hyundi Motors is building a 446 00:20:50,079 --> 00:20:54,680 dedicated facility to manufacture thirty thousand of these units annually. 447 00:20:54,480 --> 00:20:58,319 Speaker 1: Thirty thousand a year. That's not a boutique experiment. That's 448 00:20:58,359 --> 00:20:59,559 an army of workers. 449 00:20:59,599 --> 00:21:02,960 Speaker 2: They expect deployment in American factories within the next two years. 450 00:21:03,279 --> 00:21:06,000 These things will be swapping their own batteries out working 451 00:21:06,039 --> 00:21:09,519 twenty four hour shifts. The shift from cool prototype on 452 00:21:09,559 --> 00:21:13,920 YouTube to mass manufactured labor is officially here. It's happening. 453 00:21:14,000 --> 00:21:16,160 Speaker 1: Okay, we've talked about the Fighter, the Skater, and the 454 00:21:16,200 --> 00:21:19,599 factory Titan. All of them are unbelievably impressive, but they're 455 00:21:19,640 --> 00:21:23,240 also While they're metal, they're cold, they're clearly machines. 456 00:21:22,960 --> 00:21:24,799 Speaker 2: Which brings us to the other end of the spectrum, 457 00:21:24,839 --> 00:21:28,119 the psychological end, the Uncanny. 458 00:21:27,440 --> 00:21:32,240 Speaker 1: Valley Droid up This Shanghai startup unveiled a robot called Moya, 459 00:21:32,680 --> 00:21:34,880 And this one isn't about how much it can lift 460 00:21:34,960 --> 00:21:37,400 or how hard it can kick. It's about how it feels. 461 00:21:37,720 --> 00:21:40,720 Speaker 2: The uncanny valley is that term we use to describe 462 00:21:41,000 --> 00:21:44,599 that creepy, unsettling feeling when a robot looks almost human 463 00:21:45,200 --> 00:21:48,319 but not quite. There's just something wrong about it. Moya 464 00:21:48,400 --> 00:21:50,160 is trying to bridge that valley by. 465 00:21:50,160 --> 00:21:52,640 Speaker 1: Being warm, literally warm to the touch. 466 00:21:52,839 --> 00:21:56,039 Speaker 2: Yes, the skin is a very high end silicone, but 467 00:21:56,119 --> 00:21:59,319 it's heated from within. It mimics human skin texture and 468 00:21:59,359 --> 00:22:00,000 thermal output. 469 00:22:00,319 --> 00:22:02,599 Speaker 1: I have to stop you there because the idea of 470 00:22:02,680 --> 00:22:07,000 shaking a robot's hand and feeling heat that just messes 471 00:22:07,039 --> 00:22:10,880 with my head. That feels deeply, deeply weird, maybe even deceptive. 472 00:22:11,039 --> 00:22:14,200 Speaker 2: It's a powerful psychological trigger. Evolution is taught us for 473 00:22:14,279 --> 00:22:17,200 millions of years that warm things are alive, right, and 474 00:22:17,240 --> 00:22:20,160 cold things are dead or inanimate. When you touch something 475 00:22:20,200 --> 00:22:23,640 that looks human and feels warm, your brain instantly categorizes 476 00:22:23,680 --> 00:22:26,079 it differently. It creates a sense of presence, a sense 477 00:22:26,079 --> 00:22:28,839 of life. It basically hacks your empathy response. 478 00:22:29,039 --> 00:22:30,599 Speaker 1: And the face is just as advanced. 479 00:22:30,599 --> 00:22:33,839 Speaker 2: Irate twenty five degrees of freedom in the face alone. 480 00:22:33,519 --> 00:22:36,039 Speaker 1: Explain degrees of freedom for the listeners. What does that mean. 481 00:22:36,279 --> 00:22:40,000 Speaker 2: It means there are twenty five separate tiny motors or 482 00:22:40,039 --> 00:22:44,559 actuators just controlling the facial movements. This allows for incredibly 483 00:22:44,599 --> 00:22:46,000 subtle micro expressions. 484 00:22:46,160 --> 00:22:48,880 Speaker 1: So not just a fixed smile or a frown. 485 00:22:48,880 --> 00:22:53,519 Speaker 2: No, a slight, almost imperceptible raise of one eyebrow, a 486 00:22:53,559 --> 00:22:57,599 subtle purse of the lips, the tiny nonverbal cues that 487 00:22:57,640 --> 00:23:01,359 we humans use to communicate complex emotion without even saying 488 00:23:01,400 --> 00:23:01,799 a word. 489 00:23:02,000 --> 00:23:05,880 Speaker 1: And physically, Moya is much smaller than something like Atlas. 490 00:23:05,480 --> 00:23:07,559 Speaker 2: A way smaller. It's five foot five hours, so about 491 00:23:07,599 --> 00:23:10,759 average human height, and weighs only thirty two kilograms. 492 00:23:10,319 --> 00:23:14,240 Speaker 1: Thirty two kilogram that's like seventy pounds. That is incredibly light, 493 00:23:14,559 --> 00:23:15,480 shockingly light. 494 00:23:15,759 --> 00:23:18,599 Speaker 2: It uses a walker skeleton, which they actually evolved from 495 00:23:18,640 --> 00:23:21,039 a previous bot they built that was designed to run marathons, 496 00:23:21,400 --> 00:23:24,720 and it uses these lightweight lattice muscle materials instead of heavy, 497 00:23:24,720 --> 00:23:27,640 bulky motors for everything. It's designed to be non threatening. 498 00:23:27,440 --> 00:23:29,480 Speaker 1: And what is the use case here? It's not fighting 499 00:23:29,480 --> 00:23:32,119 in the octagon or working in a factory definitely not. 500 00:23:32,720 --> 00:23:37,000 Speaker 2: This is designed for elder care, for companionship. If you 501 00:23:37,039 --> 00:23:39,359 are building a robot to help an elderly person out 502 00:23:39,359 --> 00:23:41,559 of a chair or just to sit with them and 503 00:23:41,599 --> 00:23:44,759 provide a comforting presence, you don't want a seven point 504 00:23:44,799 --> 00:23:46,359 five foot industrial giant. 505 00:23:46,960 --> 00:23:48,319 Speaker 1: No, that would be terrifying. 506 00:23:48,400 --> 00:23:51,759 Speaker 2: You want something approachable, something soft, something warm. 507 00:23:52,119 --> 00:23:55,119 Speaker 1: It's fascinating, but it also raises that question you mentioned earlier. 508 00:23:55,319 --> 00:23:58,160 If it looks human, fuels warm, and reacts with emotion, 509 00:23:59,319 --> 00:24:01,359 does that make make us accept it more or does 510 00:24:01,400 --> 00:24:03,799 it make it harder to remember that it's still just 511 00:24:03,839 --> 00:24:04,640 a machine. 512 00:24:05,160 --> 00:24:08,599 Speaker 2: That is the ethical tightrope they're walking. We project emotions 513 00:24:08,640 --> 00:24:11,559 onto our cars and our toasters. Imagine what we will 514 00:24:11,599 --> 00:24:12,599 project on Da Mooya. 515 00:24:13,000 --> 00:24:14,759 Speaker 1: That's a really good point, but you know, not every 516 00:24:14,839 --> 00:24:17,200 robot is perfect, and honestly, after hearing about all this 517 00:24:17,279 --> 00:24:19,640 incredible success, I think that's a good thing. We need 518 00:24:19,680 --> 00:24:22,000 to talk about Xpang in their robot Iron. 519 00:24:22,279 --> 00:24:23,880 Speaker 2: Ah. Yes, yeah. 520 00:24:23,920 --> 00:24:26,960 Speaker 1: The Shenzen incident, the face plant heard around the world. 521 00:24:27,119 --> 00:24:30,279 Speaker 2: It was a very very public failure. Expang's Iron robot 522 00:24:30,640 --> 00:24:33,440 was doing a demo and a crowded shopping mall in Chenzen. 523 00:24:33,599 --> 00:24:34,880 It was just walking through the crowd. 524 00:24:34,960 --> 00:24:36,480 Speaker 1: Yeah, it just face planted. 525 00:24:36,359 --> 00:24:39,559 Speaker 2: Lost its balance and went down hard face first. The 526 00:24:39,640 --> 00:24:41,119 video was everywhere. 527 00:24:40,640 --> 00:24:44,200 Speaker 1: Instantly, but the CEO, a guy named Hang, he didn't 528 00:24:44,240 --> 00:24:47,000 try to bury it. He didn't issue copyright strikes on 529 00:24:47,039 --> 00:24:48,640 the videos or pretend it didn't happen. 530 00:24:48,799 --> 00:24:51,400 Speaker 2: No, and that was a very very smart move. He 531 00:24:51,440 --> 00:24:53,160 got out in front of it and called it part 532 00:24:53,200 --> 00:24:56,440 of the learning process. He embraced the failure publicly. 533 00:24:56,759 --> 00:24:58,960 Speaker 1: There was also a controversy that the robot was fake. 534 00:24:59,079 --> 00:25:01,920 Right before it even fell, people thought it moved too smoothly. 535 00:25:02,160 --> 00:25:04,920 Speaker 2: It's the reverse of the T eight hundred skepticism. With 536 00:25:04,960 --> 00:25:07,160 the T eight hundred, people said it's too good to 537 00:25:07,160 --> 00:25:10,160 be real. With Iron, people said it's too smooth to 538 00:25:10,160 --> 00:25:12,839 be a robot. They accused x Bang of just putting 539 00:25:12,880 --> 00:25:13,799 a human in a suit. 540 00:25:13,880 --> 00:25:15,519 Speaker 1: So how did they prove it wasn't a guy named 541 00:25:15,559 --> 00:25:16,920 Steve in a latex mask. 542 00:25:17,359 --> 00:25:21,519 Speaker 2: The CEO, during a live demonstration, literally peeled back the 543 00:25:21,640 --> 00:25:26,119 robot's skin on its arm. No way, Yes, live on stage. 544 00:25:26,559 --> 00:25:30,599 He exposed the motors, the wiring, the lattice muscles underneath. 545 00:25:30,680 --> 00:25:33,680 Speaker 1: That's pretty metal. You think I'm fake? Here's my insides. 546 00:25:34,200 --> 00:25:37,759 Speaker 2: It completely silenced the critics, but it also highlighted their 547 00:25:37,839 --> 00:25:41,640 unique design philosophy. They are using these lattice muscles to 548 00:25:41,759 --> 00:25:46,240 mimic biological muscle. It gives that smooth, fluid contraction that 549 00:25:46,279 --> 00:25:49,200 looks so human, which is exactly why people were suspicious 550 00:25:49,240 --> 00:25:49,920 in the first place. 551 00:25:49,960 --> 00:25:51,920 Speaker 1: And they have a pony robot too. I saw something 552 00:25:51,960 --> 00:25:52,359 about that. 553 00:25:52,359 --> 00:25:55,799 Speaker 2: They do a quadruped so walks on four legs, but 554 00:25:55,880 --> 00:25:58,759 this one has a brilliant twist. It has a flexible 555 00:25:58,839 --> 00:26:00,640 robotic arm that acts like a tail. 556 00:26:00,880 --> 00:26:03,319 Speaker 1: A tail look a dog, but a functional tail. 557 00:26:03,839 --> 00:26:05,920 Speaker 2: It can tuck between its legs when it's not needed, 558 00:26:06,359 --> 00:26:08,880 or it can extend out to grab objects, open doors, 559 00:26:08,920 --> 00:26:12,400 press elevator buttons. It's a brilliant use of the quadruped 560 00:26:12,440 --> 00:26:14,759 form factor adding a fifth manipulator. 561 00:26:14,960 --> 00:26:17,039 Speaker 1: I think the biggest takeaway from XPEG for me is 562 00:26:17,039 --> 00:26:20,200 that transparency works. Seeing a robot fall and seeing the 563 00:26:20,240 --> 00:26:22,680 company own it makes me trust them more than seeing 564 00:26:22,720 --> 00:26:26,279 another perfectly edited sizzle reel where nothing ever goes wrong. 565 00:26:26,359 --> 00:26:30,039 Speaker 2: Absolutely. It reminds us that these are incredibly complex systems 566 00:26:30,319 --> 00:26:34,519 operating in a chaotic real world. Failure is data. Failure 567 00:26:34,559 --> 00:26:35,200 is how you learn. 568 00:26:35,519 --> 00:26:38,160 Speaker 1: Speaking of the real world in public perception, we have 569 00:26:38,240 --> 00:26:39,599 to talk about the Super Bowl. 570 00:26:39,519 --> 00:26:41,519 Speaker 2: The biggest advertising stage on the planet. 571 00:26:41,559 --> 00:26:45,440 Speaker 1: Open Ai they aired an AD one hundred million plus viewers, 572 00:26:45,680 --> 00:26:48,799 and everyone was expecting, I don't know, chatchept on a screen, 573 00:26:49,119 --> 00:26:50,839 some abstract sci fi voiceover. 574 00:26:51,039 --> 00:26:56,160 Speaker 2: Instead we got hands, just hands, that's it, human hands working, creating, building, painting, 575 00:26:56,640 --> 00:27:01,039 and then so subtly robotic hands joining us in being 576 00:27:01,039 --> 00:27:02,319 guided by the human hands. 577 00:27:02,440 --> 00:27:05,720 Speaker 1: It was so understated. There were no explosions, no glowing 578 00:27:05,759 --> 00:27:07,960 red terminator eyes, nothing scary at. 579 00:27:07,880 --> 00:27:10,160 Speaker 2: All, and that was the brilliance of it. It framed 580 00:27:10,279 --> 00:27:14,759 robots not as autonomous overlords or are scary replacements, but 581 00:27:14,839 --> 00:27:18,680 as tools, as extensions of human intent and creativity. 582 00:27:18,799 --> 00:27:21,880 Speaker 1: They mentioned Codex, which is their coding agent, but the 583 00:27:21,960 --> 00:27:25,480 visual language was all about physical manipulation, about touching the 584 00:27:25,519 --> 00:27:26,359 world exactly. 585 00:27:26,400 --> 00:27:29,759 Speaker 2: OpenAI is primarily a software company, right, an AI company, 586 00:27:30,119 --> 00:27:34,519 But putting hardware specifically by manual robotic manipulation in the 587 00:27:34,519 --> 00:27:37,680 most expensive AD slot on Earth is a massive signal. 588 00:27:37,759 --> 00:27:38,519 Speaker 1: What's the signal? 589 00:27:38,640 --> 00:27:40,839 Speaker 2: They are telling us that intelligence is flowing out of 590 00:27:40,839 --> 00:27:42,960 the computer screens and into the physical. 591 00:27:42,599 --> 00:27:44,519 Speaker 1: World, embodied AI. 592 00:27:44,559 --> 00:27:48,440 Speaker 2: Embodied AI the concept that to truly understand the world, 593 00:27:48,559 --> 00:27:51,480 to have common sense, AI needs a body. It needs 594 00:27:51,519 --> 00:27:54,519 to be able to touch things, move things, and learn from 595 00:27:54,519 --> 00:27:55,400 physical interaction. 596 00:27:55,759 --> 00:27:57,920 Speaker 1: And the fact that they showed it as a collaboration 597 00:27:58,200 --> 00:28:01,680 human hand guiding robot eid hand it just normalizes the 598 00:28:01,720 --> 00:28:04,359 whole idea. It says, this is just a fancy hammer, 599 00:28:04,599 --> 00:28:06,799 this is just a better screwdriver. Don't be afraid. 600 00:28:07,000 --> 00:28:10,400 Speaker 2: It effectively de escalates the fear, unlike the T eight hundred, 601 00:28:10,440 --> 00:28:13,039 which amps up the fear in the spectacle for entertainment. 602 00:28:13,480 --> 00:28:16,720 Open AI is trying to make robotics seem almost boring. 603 00:28:17,480 --> 00:28:21,480 Utilitarian and boring is how a technology becomes ubiquitous. 604 00:28:21,599 --> 00:28:23,599 Speaker 1: So if we pull this thread all the way through, 605 00:28:23,799 --> 00:28:26,400 we have covered the entire spectrum. Today we have the 606 00:28:26,519 --> 00:28:30,039 Gladiator engine AI's T eight hundred training for combat to 607 00:28:30,079 --> 00:28:33,920 become unbreakable. We have the athlete Unitares G one learning 608 00:28:33,960 --> 00:28:36,440 physics to skate through the world. We have the worker 609 00:28:36,519 --> 00:28:39,680 Boston Dynamics at LISS, the seven point five foot giant 610 00:28:39,680 --> 00:28:42,119 ready for the factory floor, the Industrial Titan. We have 611 00:28:42,160 --> 00:28:44,920 the companion Moya with its warm skin and emotional face, 612 00:28:45,000 --> 00:28:48,039 and finally we have the tool open AI's quiet vision 613 00:28:48,079 --> 00:28:49,240 of collaborative hands. 614 00:28:49,559 --> 00:28:51,960 Speaker 2: It's a completely diverging path. It's like we're at a 615 00:28:51,960 --> 00:28:54,480 fork in the road. On one side, we are actively 616 00:28:54,480 --> 00:28:57,920 building a new species of athletes and warriors, machines that 617 00:28:58,000 --> 00:29:02,319 can physically outperform us withstand punishment we can't, and fight 618 00:29:02,359 --> 00:29:04,559 in ways we can only imagine. And on the other 619 00:29:04,680 --> 00:29:08,680 on the other side, we're building these gentle warm, collaborative 620 00:29:08,720 --> 00:29:11,319 extensions of ourselves, helpers and tools. 621 00:29:11,519 --> 00:29:13,079 Speaker 1: And that's the big question I want to leave our 622 00:29:13,119 --> 00:29:16,319 listeners with today. Which of those paths is going to 623 00:29:16,359 --> 00:29:19,079 define the future? Are we building competitors or are we 624 00:29:19,160 --> 00:29:22,279 building partners? The T eight hundred is cool, there's no 625 00:29:22,319 --> 00:29:25,359 doubt about. It's thrilling to watch. But do you want 626 00:29:25,400 --> 00:29:27,519 a T eight hundred in your house? I do not, 627 00:29:27,880 --> 00:29:28,960 Or do you want Amoya? 628 00:29:29,640 --> 00:29:33,000 Speaker 2: Or perhaps the future isn't one or the other. Maybe 629 00:29:33,000 --> 00:29:35,039 it's a world where the factory floor is run by 630 00:29:35,039 --> 00:29:38,319 the giants, the streets are navigated by the skaters, and 631 00:29:38,359 --> 00:29:41,880 our homes and hospitals are shared with the companions, a 632 00:29:41,920 --> 00:29:42,960 specialized future. 633 00:29:43,079 --> 00:29:45,359 Speaker 1: It's a lot to process. I really want to ask you, 634 00:29:45,440 --> 00:29:48,359 the listener a direct question, would you step into the 635 00:29:48,440 --> 00:29:50,640 ram with a T eight hundred, even with the best 636 00:29:50,720 --> 00:29:53,240 armor on the planet. I know my answer absolutely not. 637 00:29:54,039 --> 00:29:58,279 Or maybe more realistically, how would you feel holding the 638 00:29:58,319 --> 00:30:01,480 warm hand of a robot like Moya? Would that comfort 639 00:30:01,519 --> 00:30:03,000 you or would it just creep you out? 640 00:30:03,319 --> 00:30:05,279 Speaker 2: It's a question of psychology as much as it is 641 00:30:05,279 --> 00:30:06,880 a question of technology. 642 00:30:06,359 --> 00:30:08,839 Speaker 1: Isn't it It really is. Leave a comment and let 643 00:30:08,960 --> 00:30:10,799 us know your take. We really want to hear what 644 00:30:10,839 --> 00:30:14,519 you think about this absolute explosion of humanoid tech. 645 00:30:14,640 --> 00:30:17,559 Speaker 2: It's moving fast. Don't blink or you'll miss the next 646 00:30:17,640 --> 00:30:19,000 lead forward exactly. 647 00:30:19,440 --> 00:30:21,599 Speaker 1: Thanks for listening to thrilling Threads. We'll catch you in 648 00:30:21,640 --> 00:30:22,160 the next one.