WEBVTT - The Boston Dynamics Story 0:00:04.160 --> 0:00:07.480 Get in text with technology with tech Stuff from staff 0:00:07.520 --> 0:00:13.800 works dot com. Hey there, and welcome to tech Stuff. 0:00:13.840 --> 0:00:16.840 I am your host, Jonathan Strickland. I'm an executive producer 0:00:16.880 --> 0:00:19.079 here at how Stuff Works, and I love all things tech, 0:00:19.600 --> 0:00:22.760 and today I thought I would look into a pretty 0:00:22.840 --> 0:00:27.120 cool company, one that's been around longer than I realized, 0:00:27.640 --> 0:00:29.880 and one that has created a lot of interesting stuff 0:00:29.920 --> 0:00:32.639 and yet nothing that you can actually buy. So when 0:00:32.640 --> 0:00:35.080 am I talking about? Well, a few years ago, there 0:00:35.080 --> 0:00:38.080 were some videos that came out of a company called 0:00:38.120 --> 0:00:42.720 Boston Dynamics that went absolutely viral, and those videos showed 0:00:42.880 --> 0:00:47.280 four legged robots traversing various types of terrain, sometimes while 0:00:47.360 --> 0:00:52.120 enduring some rather abusive behavior courtesy of some humans, and 0:00:52.400 --> 0:00:55.120 that kind of served as an introduction for a lot 0:00:55.160 --> 0:00:57.880 of people to this company, Boston Dynamics. But as it 0:00:57.920 --> 0:01:01.440 turns out, it had been around a lot longer than 0:01:01.600 --> 0:01:04.479 you might have anticipated based on those videos. In fact, 0:01:04.520 --> 0:01:08.480 it turned twenty five years old in seventeen. So what 0:01:08.760 --> 0:01:12.959 is the story behind the company and its robots. Well, 0:01:13.000 --> 0:01:17.080 the founder of Boston Dynamics is Mark rayber. Raber was 0:01:17.120 --> 0:01:20.679 born in nineteen forty nine and he was a studious Fellow. 0:01:21.160 --> 0:01:25.720 He attended Northeastern University from nineteen sixty eight to nineteen 0:01:25.840 --> 0:01:29.120 seventy three and earned a degree in electrical engineering. He 0:01:29.200 --> 0:01:32.479 then went on to pursue his PhD at the Massachusetts 0:01:32.520 --> 0:01:36.959 Institute of Technology better known as m i T, and 0:01:37.040 --> 0:01:40.240 upon graduating, Dr Rayber got a job at the NASA 0:01:40.319 --> 0:01:43.880 Jet Propulsion Laboratory as a member of the technacal staff. 0:01:43.959 --> 0:01:47.600 So you could say that he was a rocket scientist, 0:01:47.880 --> 0:01:51.480 although really he worked with rocket scientists. His main focus 0:01:51.520 --> 0:01:53.919 was obviously on robotics, and he worked there for about 0:01:53.920 --> 0:01:58.200 three years, leaving in nineteen eighty. His next gig was 0:01:58.680 --> 0:02:01.480 a pretty just went as well. He was an associate 0:02:01.480 --> 0:02:05.040 professor at Carnegie Melon University's Computer science department and the 0:02:05.160 --> 0:02:08.920 Robotics Institute. Now, if you've been listening to tech stuff recently, 0:02:09.000 --> 0:02:11.840 you may remember I mentioned this particular group in a 0:02:11.880 --> 0:02:17.360 recent episode about Uber, because Uber rated Carnegie Melon and 0:02:17.480 --> 0:02:21.680 hired about forty of their robotics experts over to help 0:02:21.760 --> 0:02:26.920 them with their efforts in developing autonomous car programs. Of course, 0:02:26.960 --> 0:02:30.520 by that time Dr Raber was long gone from Carnegie Melon. 0:02:30.600 --> 0:02:33.440 But I just I mentioned it because it's another link 0:02:33.520 --> 0:02:37.320 to a recent episode. In nine six, Dr rayber took 0:02:37.360 --> 0:02:41.120 a position of Professor of Electrical Engineering and Computer Science 0:02:41.160 --> 0:02:43.360 over at m I T, and there he founded the 0:02:43.480 --> 0:02:46.639 m I T Leg Lab, which sounds a bit weird, 0:02:47.480 --> 0:02:50.160 but stick with me. So this was a lab specifically 0:02:50.200 --> 0:02:55.280 dedicated to researching and designing legged robots, not leg lamps, 0:02:55.320 --> 0:02:58.560 as I was first led to believe. It was a 0:02:58.680 --> 0:03:02.920 very difficult realization to say that it wasn't a major reward. 0:03:03.000 --> 0:03:05.280 But it turns out building robots that use legs for 0:03:05.360 --> 0:03:10.679 locomotion it's really really hard. Now why is that? Why 0:03:10.720 --> 0:03:13.280 is it so difficult to build a robot that uses 0:03:13.400 --> 0:03:17.680 legs to get around? Well, first, let's consider the wheel. Now, 0:03:18.360 --> 0:03:22.760 wheels aren't really found in organisms. There are some microscopic 0:03:22.840 --> 0:03:26.680 critters that have components that are vaguely wheel and axle like, 0:03:27.480 --> 0:03:32.240 but they're not really wheels the way are various inventions 0:03:32.280 --> 0:03:35.640 have wheels. A wheel as a limb is not likely 0:03:36.120 --> 0:03:39.480 to ever evolve in nature for several reasons. For one, 0:03:39.960 --> 0:03:43.040 any mutation that would set an organism on the path 0:03:43.760 --> 0:03:47.840 figuratively speaking to developing wheels would likely at first be 0:03:47.920 --> 0:03:51.520 a negative rather than a positive mutation. So in other words, 0:03:51.720 --> 0:03:54.680 an animal with some sort of pro to wheel limb, 0:03:54.880 --> 0:03:57.560 something that's maybe like part of a wheel but not 0:03:57.640 --> 0:04:01.240 a complete wheel, probably wouldn't have very many advantages and 0:04:01.280 --> 0:04:06.240 perhaps have distinct disadvantages with regards to survival, and thus 0:04:06.360 --> 0:04:09.200 it would be unlikely that such a mutation would be 0:04:09.200 --> 0:04:13.760 passed down to future generations to evolve. So that's one thing, 0:04:13.840 --> 0:04:17.560 is that the pathway for evolution is complicated, and if 0:04:17.760 --> 0:04:21.279 you have a mutation that is more likely to get 0:04:21.320 --> 0:04:24.960 you eaten by predators, probably not going to get passed 0:04:24.960 --> 0:04:27.840 down to a lot of offspring. For another reason, the 0:04:27.960 --> 0:04:31.200 rotation of a wheel, if not due to gravity, has 0:04:31.240 --> 0:04:34.680 to come from some source of torque, that rotational power. 0:04:34.760 --> 0:04:37.120 You have to be able to provide power to a 0:04:37.120 --> 0:04:41.000 wheel to make it turn. So locomotion for multicellular organisms 0:04:41.040 --> 0:04:44.520 typically comes from muscles, right, we have our muscles that 0:04:44.560 --> 0:04:48.200 allow us to move our limbs. But there's not really 0:04:48.240 --> 0:04:51.479 a way you could affix muscles to a wheel and 0:04:51.520 --> 0:04:55.160 allow it to turn freely indefinitely. You know, just think 0:04:55.200 --> 0:04:57.760 about how your wrist turns. Like even if you had 0:04:58.360 --> 0:05:01.960 the ability to turn your wrist completely around with no 0:05:02.080 --> 0:05:05.000 pain or anything like that, you couldn't do it indefinitely. 0:05:05.040 --> 0:05:08.240 Your muscles would continuously twist until eventually they would have 0:05:08.279 --> 0:05:10.960 to untwist like a rubber band. You couldn't just keep ongoing, 0:05:11.080 --> 0:05:15.680 so that's another problem. Wheels are almost exclusively in the 0:05:15.720 --> 0:05:21.000 domain of manufactured objects. Now that being said, they're darn't efficient. 0:05:21.279 --> 0:05:25.039 They're the easiest way to create locomotion for vehicles. It 0:05:25.120 --> 0:05:27.960 is also the most energy efficient way to travel relative 0:05:28.040 --> 0:05:32.120 to speed. So it requires a relatively simple implementation to 0:05:32.200 --> 0:05:34.839 create a wheeled robot. You don't have to worry so 0:05:34.920 --> 0:05:37.039 much about degrees of freedom. You don't have to worry 0:05:37.040 --> 0:05:41.040 about pivot points, or vector controllers or balance. I mean, 0:05:41.080 --> 0:05:43.080 as long as you've designed the robot in such a 0:05:43.120 --> 0:05:46.359 way that's relatively stable on its wheels, balance is not 0:05:46.440 --> 0:05:48.080 that big of an issue. So if you're talking about 0:05:48.120 --> 0:05:51.720 the three wheeled or four wheeled robot, balance is pretty easy. 0:05:51.760 --> 0:05:53.880 Once you get down to two wheels, then things get 0:05:53.880 --> 0:05:55.920 a little more tricky. You have to build in other 0:05:56.000 --> 0:05:58.279 systems in order for it to maintain balance, but in 0:05:58.320 --> 0:06:02.279 general it's not that hard to do. They also typically 0:06:02.279 --> 0:06:05.919 cost less to develop and less to build than a 0:06:06.000 --> 0:06:09.839 legged robot. A simple three wheel robot is stable on 0:06:09.839 --> 0:06:12.560 its own. If you've got that basic sort of triangle shape, 0:06:12.760 --> 0:06:14.960 it does not require a lot of high end technology 0:06:15.000 --> 0:06:17.960 to work, just some motors and then whatever sensors and 0:06:18.080 --> 0:06:21.920 processors you need to actually control the robot. Uh. Four 0:06:21.920 --> 0:06:25.240 wheeled robots are are even more stable than three wheeled robots, 0:06:25.279 --> 0:06:27.680 and they can excel in applications that a three wheeled 0:06:27.720 --> 0:06:30.719 robot might find difficult. For example, if you plan for 0:06:30.760 --> 0:06:33.200 your robot to lift things, you could have a three 0:06:33.200 --> 0:06:36.839 wheeled robot that would become unbalanced depending upon where you 0:06:36.920 --> 0:06:39.560 put the load on top of the robot right If 0:06:39.600 --> 0:06:42.080 it's too far off to one side of its center 0:06:42.120 --> 0:06:44.360 of gravity, it can make the robots sort of tip over. 0:06:45.040 --> 0:06:47.320 Our four wheeled robot tends to be more stable in 0:06:47.320 --> 0:06:50.520 those cases. You can also design a wheeled robot to 0:06:50.520 --> 0:06:54.120 operate even if it's been flipped upside down. So if 0:06:54.160 --> 0:06:56.800 the robot is such that the wheels can make contact, 0:06:56.880 --> 0:06:59.760 whether it's right side up or upside down, you could 0:07:00.000 --> 0:07:03.360 of the option of continuously operating that robot even if 0:07:03.360 --> 0:07:05.320 it were to flip over. In fact, I remember some 0:07:05.480 --> 0:07:07.800 remote controlled cars that were designed this way. They were 0:07:07.839 --> 0:07:11.320 marketed as that you would purposefully drive your car and 0:07:11.360 --> 0:07:13.320 to say a wall, make it flip over, and then 0:07:13.320 --> 0:07:16.000 you could keep on driving the car just kind of fun. 0:07:16.520 --> 0:07:19.400 That's the same basic idea, but wheels are not ideal 0:07:19.440 --> 0:07:22.880 across all different types of terrain. They work best on 0:07:23.000 --> 0:07:26.239 tracks if you have them, but if you don't have tracks, 0:07:26.280 --> 0:07:30.480 they work best on smooth even ground. Depending upon the 0:07:30.520 --> 0:07:33.400 wheel design and the torque available, they may be able 0:07:33.480 --> 0:07:37.280 to travel over semi rough terrain too fairly rough terrain, 0:07:37.360 --> 0:07:39.640 but they aren't great for everything, and if the ground 0:07:39.680 --> 0:07:42.720 is really challenging, they can prove to be ineffective. So 0:07:42.760 --> 0:07:44.760 that challenge could be that the ground is just too 0:07:44.840 --> 0:07:48.200 rocky or covered in debris, or it's composed of loose 0:07:48.240 --> 0:07:51.640 elements like sand, and in those situations, legs might be 0:07:51.640 --> 0:07:56.400 more effective. Your wheels might otherwise spin needlessly, just completely 0:07:56.720 --> 0:08:00.000 without motion. You're just turning around and you're digging yourself 0:07:59.840 --> 0:08:03.440 into a hole. Or they could end up catching on 0:08:03.560 --> 0:08:06.320 debris and pulling them into the works of the robot, 0:08:06.520 --> 0:08:10.480 gumming everything up. A legged robot with the right control 0:08:10.520 --> 0:08:14.520 system or programming can step over obstacles and move smoothly 0:08:14.600 --> 0:08:17.400 over different types of terrain, robots can use legs to 0:08:17.440 --> 0:08:19.960 select a precise spot on the ground to place its 0:08:20.000 --> 0:08:23.600 weight and then shift to move. A wheel has little 0:08:23.680 --> 0:08:26.480 choice but to just roll ahead, but a legged robot 0:08:26.520 --> 0:08:28.720 can place its legs and then shift its weight in 0:08:28.760 --> 0:08:31.800 a very specific way, which is helpful on uneven ground 0:08:31.920 --> 0:08:33.920 or going up and down the stairs. And depending upon 0:08:33.920 --> 0:08:36.319 the design of the robot, you can make it where 0:08:36.880 --> 0:08:39.440 the way it places its leg down, the way it 0:08:39.480 --> 0:08:42.559 shifts its weight, it can maintain its relative position over 0:08:42.600 --> 0:08:44.320 the ground. So it looks like if you if you 0:08:44.360 --> 0:08:47.640 were to remove the legs magically from the image, like 0:08:47.679 --> 0:08:51.600 it's just floating over the landscape. But it's all because 0:08:51.679 --> 0:08:53.679 the way it's able to place its legs and move 0:08:53.720 --> 0:08:57.360 its weight around. But legs, as it turns out, are 0:08:57.400 --> 0:09:01.559 super hard to design. They require joints and points of articulation, 0:09:01.920 --> 0:09:04.000 and you have to have power to move everything and 0:09:04.040 --> 0:09:07.760 figure out how you're going to create the force necessary 0:09:07.800 --> 0:09:10.840 to open and close those various joints, and you have 0:09:10.880 --> 0:09:13.680 to have lots of precise directions to work effectively. So 0:09:13.720 --> 0:09:16.840 you need a very efficient computer and you need really 0:09:16.840 --> 0:09:20.240 good programming so that it can actually maneuver over these 0:09:20.280 --> 0:09:23.679 different terrain. A leg needs to have more moving parts 0:09:23.720 --> 0:09:26.040 than a wheel, with at least three actuators to make 0:09:26.080 --> 0:09:28.440 the leg more useful than a wheel. You can have 0:09:28.520 --> 0:09:30.959 fewer than three actuators and have a leg, but it's 0:09:30.960 --> 0:09:32.679 not going to be more useful than a wheel unless 0:09:32.679 --> 0:09:36.000 you have at least three actuators. So a lot of 0:09:36.000 --> 0:09:39.880 the work in legged robots concentrates on how to move 0:09:40.000 --> 0:09:44.040 those legs. So, for example, you might use artificial muscles 0:09:44.080 --> 0:09:48.280 with stuff like electroactive polymers. Now, those are materials made 0:09:48.280 --> 0:09:51.040 out of long chain molecules that change their shape when 0:09:51.080 --> 0:09:54.240 they encounter an electric field. So here's an example, and 0:09:54.240 --> 0:09:56.000 this is just one I pulled off the top of 0:09:56.040 --> 0:09:58.600 my head, and it's not really indicative of how they 0:09:58.600 --> 0:10:01.280 all work, but it's way they at work. Let's say 0:10:01.280 --> 0:10:03.760 that you build a structure and it's made off of 0:10:03.760 --> 0:10:08.000 this material, and that structure is elongated, it's extended, but 0:10:08.040 --> 0:10:11.400 if it encounters an electric field, it contracts, it pulls together. 0:10:11.880 --> 0:10:14.040 This would make it work in a way similar to 0:10:14.080 --> 0:10:18.480 the way our muscles work. Now, there's not just that 0:10:18.600 --> 0:10:21.160 kind of method to power legs. In fact, that's a 0:10:21.160 --> 0:10:25.280 pretty rare one. You're more likely to find electric actuators 0:10:25.360 --> 0:10:29.319 or hydraulic actuators. But that's another challenge is how do 0:10:29.360 --> 0:10:32.160 you actually get the legs to move. Then there's a 0:10:32.200 --> 0:10:34.680 ton of work on that processing side to calculate things 0:10:34.679 --> 0:10:38.959 like balanced momentum, weight distribution, path finding, all of these 0:10:39.000 --> 0:10:42.640 different things that we take for granted we learn it intuitively, 0:10:43.120 --> 0:10:47.080 but that doesn't work with robots necessarily. And there's a 0:10:47.120 --> 0:10:50.199 long history of studies that happened before robotics were even 0:10:50.240 --> 0:10:54.360 a thing that would eventually inform the design of legged robots, 0:10:54.360 --> 0:10:57.600 and some of those studies which went on to help 0:10:57.720 --> 0:11:00.960 future engineers when they were building out these robot legs 0:11:01.000 --> 0:11:04.360 for the first time. We're pretty grim, as it turns out. 0:11:04.760 --> 0:11:09.439 So get ready for a gross story that I'm about 0:11:09.440 --> 0:11:13.320 to tell you. One study that the Leg Lab cites 0:11:13.400 --> 0:11:16.720 as being an important milestone in the development of legged 0:11:16.840 --> 0:11:21.960 robots comes from an eighteen thirty six journal and it 0:11:22.000 --> 0:11:27.040 involves dead people. The title of this study was Mechanic 0:11:27.360 --> 0:11:33.079 derminsch lichen I and a Thomish physiologic unter schung fondap 0:11:33.120 --> 0:11:38.040 rudin Wilhelm Weber w Edward web Webber. All right, so 0:11:38.120 --> 0:11:40.280 translated to English because that was a terrible German, I know, 0:11:40.360 --> 0:11:44.840 my German's awful kind of Deutsches. It turns out it 0:11:44.920 --> 0:11:49.040 actually means the mechanics of the human and anatomical physiological 0:11:49.080 --> 0:11:53.120 examination of the brothers Wilhelm and Edward Webber. Now their 0:11:53.120 --> 0:11:58.040 work is available to read if you read German. But 0:11:58.200 --> 0:11:59.760 they found that if you took the leg of a 0:12:00.000 --> 0:12:04.360 corps and you swung the leg of the corpse, so 0:12:04.440 --> 0:12:08.920 it acted like a compound pendulum. The swinging motion was 0:12:09.040 --> 0:12:10.959 very similar to the cadence of a leg with a 0:12:10.960 --> 0:12:16.880 live person is walking around. That's cheerful. But as it 0:12:16.920 --> 0:12:19.640 turns out, that would actually be one of those fundamental 0:12:19.720 --> 0:12:22.800 studies that would go on to inform people who were 0:12:22.840 --> 0:12:27.120 trying to build artificial legs and mechanical legs. Most of 0:12:27.120 --> 0:12:30.560 the studies the lab sites aren't nearly so grim and 0:12:30.640 --> 0:12:35.040 don't involve nearly as many corpses. So for example, another 0:12:35.080 --> 0:12:38.800 important contribution came from a guy named Edward Moybridge. Now 0:12:38.840 --> 0:12:42.400 this was an English photographer who developed stop motion photography. 0:12:42.720 --> 0:12:46.199 Moybridge was hired by a fat cat named Leland Stanford. 0:12:46.200 --> 0:12:49.600 He was a former governor and a railroad tycoon and 0:12:49.640 --> 0:12:53.040 Stanford had this idea, and he needed evidence to support 0:12:53.080 --> 0:12:56.760 his idea. His idea was that when a horse is 0:12:56.800 --> 0:13:00.800 in full gallop, at some point during its gait, it 0:13:00.840 --> 0:13:05.120 will have all four hoofs off the ground simultaneously, and 0:13:05.240 --> 0:13:10.480 moy Bridge was essentially contracted to prove that Stanford had 0:13:10.520 --> 0:13:13.320 a leg to stand on as it were. So Moybridge 0:13:13.360 --> 0:13:16.320 invented away to use a series of twelve cameras to 0:13:16.400 --> 0:13:19.200 take photos with an exposure lasting only a fraction of 0:13:19.200 --> 0:13:21.800 a second, which at the time it was pretty remarkable. 0:13:21.880 --> 0:13:26.280 Most photo photography. Early photography had very long exposure time 0:13:26.400 --> 0:13:29.320 in order to get enough light to actually create the effect, 0:13:29.640 --> 0:13:32.280 which is why you have lots of portraits of very 0:13:32.400 --> 0:13:35.720 dour looking people in early photography, because it meant they 0:13:35.760 --> 0:13:39.319 had to sit still for several minutes at a time, 0:13:39.440 --> 0:13:41.440 and you know, if you're trying to hold a smile, 0:13:42.360 --> 0:13:44.320 pretty soon you hate the world and everything in it, 0:13:44.400 --> 0:13:47.200 so you typically have these very kind of neutral or 0:13:47.240 --> 0:13:51.120 dour expressions in those early photographs. Well, Moybridge wanted to 0:13:51.120 --> 0:13:53.000 find a way of doing this much faster, so he 0:13:53.080 --> 0:13:55.839 created this camera system where in a fraction of a 0:13:55.880 --> 0:13:58.400 second it could expose film to enough light to create 0:13:58.400 --> 0:14:01.319 an image. He said, twelve of these cameras in a row, 0:14:01.880 --> 0:14:04.680 and then he used little trip wires not to trip 0:14:04.760 --> 0:14:07.000 up the horses, but rather the wires themselves would trip 0:14:07.040 --> 0:14:09.880 when a horse ran through them, and it would cause 0:14:09.920 --> 0:14:13.520 the cameras to take these photos in sequence, and then 0:14:13.559 --> 0:14:16.480 you could look at them and see a horse's gallop 0:14:16.760 --> 0:14:20.480 in progress. The photos showed that Stanford was in fact correct. 0:14:20.600 --> 0:14:22.680 There was a brief moment in the horse's stride in 0:14:22.720 --> 0:14:25.760 which all four legs were off the ground, and Moybridge's 0:14:25.800 --> 0:14:29.520 innovation became valuable for scientists who wanted to study animal movement, 0:14:29.960 --> 0:14:33.240 and later for engineers and roboticists who wished to mimic 0:14:33.480 --> 0:14:38.040 those movements when they were developing robotic legs. Uh, Moybridge 0:14:38.080 --> 0:14:41.160 and stop motion animation and stop motion photography, I should say, 0:14:41.200 --> 0:14:44.000 are really fascinating subjects, and I'm sure I'll cover them 0:14:44.000 --> 0:14:47.160 in another episode of tech Stuff at some point. Back 0:14:47.200 --> 0:14:50.240 to milestones, Well, there are a lot of others that 0:14:50.280 --> 0:14:52.760 the lab acknowledgeist as being really important moments in the 0:14:52.760 --> 0:14:55.880 development of robotic legs, and they even include a few 0:14:55.920 --> 0:14:59.280 of Dr Rabur's contributions to the fields, such as three. 0:14:59.280 --> 0:15:03.400 When Rayber build a one legged hopping machine, it could 0:15:03.440 --> 0:15:06.000 keep its own balance and travel at a specified rate. 0:15:06.040 --> 0:15:08.920 It kind of looks like a UFO on a pogo stick. 0:15:09.000 --> 0:15:13.520 It's constantly bouncing around. The main trick here was just 0:15:13.600 --> 0:15:17.160 getting that balance right. So Rayber was able to create 0:15:17.240 --> 0:15:20.840 this robot that uses a computer that calculates where that 0:15:21.040 --> 0:15:23.960 one foot needs to come down in order to keep 0:15:24.000 --> 0:15:27.680 the overall robot balanced while still traveling in the desired direction. 0:15:28.560 --> 0:15:31.120 Those calculations had to be done fast enough so that 0:15:31.200 --> 0:15:33.760 they were complete before the robot would come crashing down 0:15:33.840 --> 0:15:36.400 after its last hops. So as it hops up in 0:15:36.400 --> 0:15:38.720 the air, the computer has to figure out the trajectory, 0:15:38.840 --> 0:15:42.320 the momentum, all of these different elements that the robot 0:15:42.400 --> 0:15:46.680 is experiencing in order to compensate with the leg so 0:15:46.760 --> 0:15:48.600 that it can hit the ground at just the right 0:15:48.640 --> 0:15:50.720 spot with just the right amount of force to propel 0:15:50.760 --> 0:15:53.720 the robot up again and continue moving it in whatever 0:15:53.760 --> 0:15:57.080 direction you wanted to go in. So it's actually pretty complicated, 0:15:57.120 --> 0:15:59.720 It took a lot of work and it's pretty fascinating 0:15:59.720 --> 0:16:03.560 stuff off. That basic principle would then later be adapted 0:16:03.560 --> 0:16:07.160 and applied into multi legged robots with a computer determining 0:16:07.200 --> 0:16:09.720 where each leg needs to go, how it needs to 0:16:09.760 --> 0:16:13.120 make contact with the ground, how much force, all just 0:16:13.160 --> 0:16:16.920 so that the robot doesn't fall over. Rayber would end 0:16:16.960 --> 0:16:20.760 up writing a book about his research. In the book 0:16:20.800 --> 0:16:25.520 has the title legged robots that Balance. Pretty much sums 0:16:25.600 --> 0:16:28.240 up the whole thing, doesn't it. It sounds like it's 0:16:28.240 --> 0:16:30.960 a trivial thing. But again, it was really tricky to 0:16:31.040 --> 0:16:35.200 do because while we're growing up, most of us, you know, 0:16:35.240 --> 0:16:38.680 as kids, we learned to walk, and we kind of 0:16:39.000 --> 0:16:42.040 have an kind of intuitive grasp of what we need 0:16:42.080 --> 0:16:45.240 to do in order to keep our balance barring any 0:16:45.920 --> 0:16:50.400 unforeseen circumstances like stepping on something that is on uh 0:16:50.840 --> 0:16:54.040 it's like uncertain ground without even realizing it, or getting 0:16:54.080 --> 0:16:57.240 shoved by someone. We're pretty good at keeping our balance 0:16:57.360 --> 0:17:01.600 generally speaking, assuming all things are you know, normal, But 0:17:01.640 --> 0:17:06.360 a robot has to think air quotes, think about how 0:17:06.400 --> 0:17:09.080 to do this. Intuition isn't an option, at least not 0:17:09.160 --> 0:17:13.080 without some really advanced machine learning algorithms, which honestly, we're 0:17:13.119 --> 0:17:14.960 not that far along. In the nineteen eighties, this was 0:17:15.000 --> 0:17:17.879 all stuff that had to be worked out by human engineers. 0:17:18.359 --> 0:17:21.880 These days if you created a really advanced machine learning algorithm, 0:17:22.000 --> 0:17:25.080 I suppose it won't be long before we start seeing 0:17:25.160 --> 0:17:27.720 robots that teach themselves how they need to walk. And 0:17:27.800 --> 0:17:31.000 we've seen some examples of that, and they look really weird, 0:17:32.119 --> 0:17:35.159 but these are mostly simulations, like we've seen computer simulations 0:17:35.160 --> 0:17:37.760 of what it would look like, where there's a virtual 0:17:37.880 --> 0:17:43.199 robot that has virtual statistics including its virtual weight and 0:17:43.359 --> 0:17:45.359 height and all these sort of things that have real 0:17:45.440 --> 0:17:49.200 world impact if you were to create an actual robot, 0:17:49.560 --> 0:17:51.960 And then there are simulations that show how a computer 0:17:52.040 --> 0:17:54.320 works out how this robot needs to move in order 0:17:54.359 --> 0:17:57.640 to actually walk without falling over, and in some cases 0:17:57.960 --> 0:18:02.280 the animations are hysterical and also nightmare inducing. But as 0:18:02.320 --> 0:18:04.680 far as I know, we don't have any actual physical 0:18:04.800 --> 0:18:07.359 robots that follow that model. I could be wrong about that. 0:18:07.400 --> 0:18:10.160 By the way, there's advances in robotics all the time, 0:18:10.280 --> 0:18:13.040 and they're very well. Maybe a project out there that's 0:18:13.119 --> 0:18:15.840 much further along than what I'm suggesting, but the ones 0:18:16.000 --> 0:18:20.159 I have seen of all been virtual representations rabs. Robots. 0:18:20.160 --> 0:18:24.440 By the way, we're mostly dynamic robots, meaning they were 0:18:24.520 --> 0:18:27.120 dynamic in the sense they needed to move in order 0:18:27.200 --> 0:18:32.760 to remain stable. You could have static stable robots, but 0:18:32.800 --> 0:18:34.960 these were ones that needed to move in order to 0:18:35.200 --> 0:18:39.240 maintain that upright position and not just fall over. And 0:18:39.280 --> 0:18:42.680 even some of those early robots, which often were tethered 0:18:42.920 --> 0:18:47.159 to control and power systems, could do really amazing things. 0:18:47.280 --> 0:18:51.280 So most of the early robots, if you see the 0:18:51.320 --> 0:18:54.480 footage from Boston Dynamics, they have all these different cables 0:18:54.480 --> 0:18:57.360 and stuff coming off of them. Those cables are frequently 0:18:57.400 --> 0:19:00.880 going to power systems and to controls the dumbs. Because 0:19:00.920 --> 0:19:04.840 the robots were meant to demonstrate specific types of locomotion, 0:19:04.880 --> 0:19:07.520 they weren't going to be used as a finished product. 0:19:08.040 --> 0:19:10.679 It was more about studying the way to build the 0:19:10.760 --> 0:19:14.639 best systems, and so you didn't have to worry about 0:19:14.680 --> 0:19:16.920 making something that was all self contained. You were studying 0:19:16.920 --> 0:19:19.199 all of this in the lab anyway. So a lot 0:19:19.240 --> 0:19:21.480 of those videos you'll see just have the this like 0:19:21.680 --> 0:19:26.040 nest of chords that are grouped up over the robot 0:19:26.119 --> 0:19:29.040 and go into some or you know, system or other 0:19:29.240 --> 0:19:34.000 multiple systems in some cases. But even in those early robots, 0:19:34.040 --> 0:19:38.440 there's some video clips of phenomenal robotic gymnastics, Like there's 0:19:38.440 --> 0:19:40.879 somewhere engineers are controlling the robots and making them do 0:19:40.960 --> 0:19:44.160 things like forward flips while traveling a path. I'm talking 0:19:44.160 --> 0:19:46.639 about like hopping robots. They'll be hopping in a circle 0:19:47.119 --> 0:19:50.520 and then they'll do an extra tall hop and do 0:19:50.600 --> 0:19:53.800 a full forward flip and land on their foot and 0:19:53.920 --> 0:19:57.120 catch their balance and continue going in the circle. This 0:19:57.160 --> 0:19:59.920 is decades before we saw the viral clips of youtu 0:20:00.160 --> 0:20:03.879 both Big Dog and the other famous robots from Boston Dynamics. 0:20:03.880 --> 0:20:07.239 So if you get a chance, do some digging on 0:20:07.280 --> 0:20:09.840 YouTube for Boston Dynamics and take a look at some 0:20:09.880 --> 0:20:13.440 of those early videos, because they're really really impressive and 0:20:13.560 --> 0:20:18.439 in some cases extremely creepy. Another Raybors contributions on the 0:20:18.480 --> 0:20:22.480 list dates from when he and his team developed quadruped 0:20:22.600 --> 0:20:26.480 robots that can move at different speeds like trotting, pacing, 0:20:26.560 --> 0:20:29.639 and bounding, and they could switch between those different modes 0:20:29.720 --> 0:20:31.240 so you didn't just have to turn it on and 0:20:31.240 --> 0:20:33.640 then put it down on the ground. You could actually 0:20:33.800 --> 0:20:36.879 go from one speed to the other and you know, 0:20:36.960 --> 0:20:39.520 knock up a notch or down a notch. If you 0:20:39.560 --> 0:20:42.479 watch slow motion footage of animals moving at different speeds, 0:20:42.480 --> 0:20:45.720 you'll see that their legs move differently at each rate 0:20:45.720 --> 0:20:48.440 of travel. So like a horse at a trot and 0:20:48.440 --> 0:20:50.679 a horse at a gallop, their legs are behaving in 0:20:50.720 --> 0:20:53.200 different ways. Well, Raybor was working very hard to create 0:20:53.280 --> 0:20:56.120 the same sort of thing in his robots. So Rayber 0:20:56.200 --> 0:20:58.800 and his team worked to push the envelope in the 0:20:58.800 --> 0:21:02.440 field of robotic low emotion. They experimented with different designs 0:21:02.520 --> 0:21:05.440 create more stable, capable robots, and many of those early 0:21:05.480 --> 0:21:09.040 designs used humans to guide the path of the robots, 0:21:09.480 --> 0:21:13.440 so it was kind of a partly human controlled, partly 0:21:13.560 --> 0:21:19.040 robotic controlled system. Frequently, the robots themselves had computers, either 0:21:19.160 --> 0:21:21.600 on board or just off board that did all the 0:21:21.640 --> 0:21:24.800 actual work to determine the particulars like where does the 0:21:24.880 --> 0:21:28.080 leg go and how would it need to shift its 0:21:28.119 --> 0:21:30.560 weight in order to obey the commands. So the human 0:21:30.640 --> 0:21:32.960 is essentially giving commands like I want you to move 0:21:33.040 --> 0:21:36.960 over forward five feet, but it was the robots computer 0:21:37.080 --> 0:21:38.960 that was actually doing the work of how does it 0:21:39.119 --> 0:21:43.399 achieve that goal. In n Rayber founded a company to 0:21:43.440 --> 0:21:47.119 continue this work and find commercial applications for the technologies 0:21:47.119 --> 0:21:51.639 he was developing, and that company was Boston Dynamics. So 0:21:51.680 --> 0:21:54.800 we're finally at the point where Boston Dynamics is its 0:21:54.840 --> 0:21:58.200 own company, which means it's time to take a quick 0:21:58.200 --> 0:22:07.600 break and thank our sponsor. All right, So it's the 0:22:07.600 --> 0:22:10.679 early nine nineties and bust and Dynamics as a new company. 0:22:10.760 --> 0:22:14.760 So do they start turning out commercial robots right away? Nope. 0:22:14.960 --> 0:22:16.600 As a matter of fact, as of the recording of 0:22:16.640 --> 0:22:19.400 this podcast, none of the company's work has ever been 0:22:19.400 --> 0:22:23.280 turned into a commercial product. Directly, you cannot go out 0:22:23.320 --> 0:22:26.399 and buy a Boston Dynamics robot for yourself. But the 0:22:26.400 --> 0:22:29.159 company's work has been extremely important to push forward the 0:22:29.200 --> 0:22:32.560 evolution of robotics, and there are some really big organizations 0:22:32.560 --> 0:22:35.760 that have been extremely interested in that, such as the 0:22:35.800 --> 0:22:39.720 Department of Defense. Now, much of Buston Dynamics funding has 0:22:39.760 --> 0:22:43.000 come from really big contracts awarded to the company by 0:22:43.040 --> 0:22:47.760 the Defense Advanced Research Projects Agency, also known as DARPA. 0:22:47.920 --> 0:22:50.160 And you might remember DARPA as being the group responsible 0:22:50.200 --> 0:22:53.040