AI transcript
0:00:21 because I didn’t know that sports biomechanics could be a career, I decided to go to grad
0:00:24 school and initially start working on prosthetic limbs.
0:00:25 Okay.
0:00:29 So that was the first two years of grad school.
0:00:35 Then about two years in, I discovered baseball pitching biomechanics research.
0:00:40 The funny thing that happened there was I gave a Ph.D. committee meeting where I spent
0:00:45 most of the time talking about prosthetic limbs and then I spent the last few minutes
0:00:51 as an aside on the baseball pitching research I found and my committee was like, “Jimmy,
0:00:55 the last few minutes were way better than the first 45.”
0:00:57 I was like, “All right.”
0:01:02 So then they were like, “We’ll let you do baseball pitching biomechanics as your Ph.D.
0:01:08 work, but just so you know, it might be really hard to have a career doing that.”
0:01:10 But they were like, “If you want to go for it, you can go for it.”
0:01:19 And so I was like, “You know what, sure, I’ll go for it.”
0:01:22 I’m Jacob Goldstein and this is What’s Your Problem.
0:01:27 Today we have the second episode in our series about people who are working at the frontiers
0:01:31 of technology to help elite athletes perform better.
0:01:34 My guest today is Jimmy Buffy.
0:01:39 And as it happened, the concerns of his grad school advisors were unfounded.
0:01:44 Jimmy has in fact made a career out of the biomechanics of pitching in baseball and sports
0:01:47 biomechanics more broadly.
0:01:51 When he finished grad school, he got a job with the Los Angeles Dodgers and he went on
0:01:56 to co-found a company called Reboot Motion that works with teams in Major League Baseball
0:01:57 and the NBA.
0:01:59 Jimmy’s problem is this.
0:02:05 How do you take massive amounts of data about how professional athletes move and turn all
0:02:12 that data into information that actually helps those athletes perform better?
0:02:17 You end up doing your dissertation research on the biomechanics of pitching, of baseball,
0:02:20 pitching in baseball.
0:02:23 And then you get hired by the Dodgers.
0:02:24 Yeah.
0:02:25 Yeah.
0:02:26 That was awesome.
0:02:29 Because I originally didn’t, again, didn’t realize that that could be a thing that could
0:02:30 happen.
0:02:32 Well, and it kind of wasn’t, right?
0:02:37 Like you’re kind of just coming into this field as it’s becoming a field where you can
0:02:39 get a job, where it’s a field, right?
0:02:40 Right, exactly.
0:02:41 Yeah.
0:02:44 I mean, the challenge then was there wasn’t a lot of options for actually even getting
0:02:47 the data that you need to analyze.
0:02:53 So 10 years ago, like what is the state of play in this sort of nascent field that you’re
0:02:56 in helping to create?
0:03:00 So the field, I would say, is like sports biomechanics.
0:03:06 And what that is, is being able to analyze the movement of athletes for lots of purposes,
0:03:10 help them reduce injury risk, help them improve performance.
0:03:15 And to be clear, like folk sports biomechanics has been around forever, right?
0:03:16 That’s what coaches do.
0:03:17 They stand there.
0:03:18 Yeah.
0:03:21 And so it’s kind of becoming, it’s becoming more technical, right?
0:03:22 The field is becoming more technical.
0:03:23 Yeah.
0:03:31 So the state of the art relied on what is called marker-based motion capture, which is where
0:03:36 you literally put reflective markers, like little balls, you stick them all over somebody’s
0:03:37 body.
0:03:41 Usually, the person has to like strip their clothes off because you want the markers like
0:03:47 literally like on the skin, on the joints, and then you have these special cameras that
0:03:48 track those markers.
0:03:53 So you’ve got like a pitcher in his underwear with a bunch of little metal balls, and they’re
0:03:55 like, just pitch like you always pitch.
0:03:56 Right.
0:03:57 And that’s the challenge.
0:04:02 That’s why it wasn’t very widespread as a thing people did because it was so hard to
0:04:07 collect the data because ultimately what you would need is you need the data on how someone
0:04:08 is moving.
0:04:12 You need to track where their elbow is, where their wrist is, where their knees are so that
0:04:13 you can analyze it.
0:04:17 And the state of the art for tracking it was an awful experience for the people you were
0:04:18 trying to track.
0:04:22 And that presumably would mean they didn’t pitch like they usually pitch.
0:04:23 Exactly.
0:04:25 Because they don’t usually stand there in their underwear with metal.
0:04:26 Were they really in their underwear?
0:04:29 By the way, I’m saying it because it sounds absurd, but is that actually what they were
0:04:30 doing?
0:04:34 That’s actually, you strip down to your boxers, your boxer briefs, and that’s it.
0:04:35 That’s all you’re wearing.
0:04:39 So it basically didn’t work and it basically wasn’t very widely used as a result.
0:04:40 Right.
0:04:41 Exactly.
0:04:47 Yeah, you look at studies in that field and people would be throwing like several miles
0:04:52 an hour slower than they would be throwing when they weren’t wearing all that stuff.
0:04:54 So what changes?
0:04:56 Computer vision.
0:04:59 That was the big inflection point.
0:05:06 Now to be fair, even when I was finishing my PhD, and I’ll give them a shout out, there
0:05:12 was a company that was already working really hard to solve this problem for baseball teams
0:05:18 that I got to be familiar with called Kinatracks.
0:05:23 But yeah, the big inflection point was computer vision, basically using artificial intelligence
0:05:31 to identify where those joints are in a camera image rather than needing to paste those reflective
0:05:32 markers.
0:05:38 So computer vision takes off, you’re working at the Dodgers, and then eventually in 2019,
0:05:43 you leave the Dodgers and you start your company reboot motion.
0:05:45 What does your company do?
0:05:49 We do what we call biomechanics as a service.
0:05:55 So we try to analyze this computer vision data at a very large scale to help teams and
0:05:58 coaches make use of it to help athletes get better.
0:05:59 Bass?
0:06:00 Bass, yeah.
0:06:05 That’s bass, but bass.
0:06:07 And who are your customers?
0:06:16 Our customers are majorly baseball teams, actually NBA teams, so we’ve gotten into basketball.
0:06:21 So sort of like league wide data providers.
0:06:24 So yeah, leagues, teams is basically our sweet spot.
0:06:30 So let’s talk in like a little more detail about what you actually do, right?
0:06:33 Tell me the story of what you do.
0:06:39 So Evan, my co-founder, Evan Demchick, he likes to call this the biomechanics train.
0:06:43 Hey, let’s take a ride on the biomechanics train.
0:06:49 And we call it that because the way our product works is we let people get on the train at
0:06:53 whatever stop works for them and get off the train at whatever stop works for them.
0:06:55 How far are we going to go with this metaphor?
0:06:57 I’m a little, I’m nervous now about it.
0:06:59 That might be as far as we go.
0:07:00 Okay.
0:07:01 Good.
0:07:02 Good.
0:07:09 So just tell it to me, start at whatever seems like the beginning of an encounter.
0:07:10 And so I’d like to understand how that works.
0:07:13 That’s really kind of a way to think about it.
0:07:16 So let’s start with the data.
0:07:19 What is the basic thing that’s happening?
0:07:27 So the very first thing that happens is you record videos of the athlete doing the athletic
0:07:28 motion.
0:07:29 So we’ll talk about pitching.
0:07:32 So you record videos of a pitcher pitching.
0:07:35 Our product, we actually have implemented our own computer vision models.
0:07:38 So we can do that if people want.
0:07:44 But generally speaking, people have systems like kinetraxx is one, I’ve mentioned that
0:07:49 Hawkeye is another popular one where there is a system in place that has the cameras
0:07:53 that record the videos and then runs those computer vision models.
0:07:58 And what those computer vision models do is they extract the locations in three dimensional
0:08:01 space of all of like the joint centers.
0:08:02 So where’s my elbow?
0:08:03 Where’s my wrist?
0:08:04 Where’s my knee?
0:08:06 In three dimensional space.
0:08:09 That’s what comes out of these computer vision systems.
0:08:12 So pretty much everybody has that at this point?
0:08:17 Like every professional baseball team has that for every pitch in every game at this
0:08:18 point?
0:08:19 Yes, exactly.
0:08:21 So it’s a ton, a ton of data.
0:08:25 So that’s another challenge that we’ve solved is not only how do you do this, but how do
0:08:28 you do this at a very large scale?
0:08:30 So this data, everybody’s got it.
0:08:35 And you’re not, you can sort of process the data, but that’s not your special sauce.
0:08:36 That’s not your secret sauce, right?
0:08:40 So typically they’ll send you that data of like, here’s all the body points, here’s how
0:08:43 they’re moving in physical space.
0:08:44 Then what do you do with it?
0:08:47 So yeah, this is where the special sauce comes in.
0:08:52 So the first step to that is how do you turn those key points into a human skeleton?
0:08:55 So you got to figure out like, where do the bones connect?
0:08:58 What sort of degrees of freedom do those bones have?
0:09:05 So then you can figure out how do those key points animate a human skeleton?
0:09:06 So you’re sort of rebuilding it.
0:09:10 It’s like you start with a picture of a person and then you turn it into a bunch of data
0:09:11 points.
0:09:13 And then you got to kind of build the person back up again from the database.
0:09:14 Exactly, exactly.
0:09:20 So then, so now you have an actual human skeleton where the shoulder is rotating, the elbow
0:09:24 is flexing, the knee is flexing, the hips are rotating.
0:09:28 And now once you do that, now you can understand that data in the context of how the body
0:09:29 works.
0:09:30 Okay.
0:09:35 So once we’ve done that, then we calculate how energy flows through the body.
0:09:37 We calculate how momentum flows through the body.
0:09:41 And once we’ve done that, we can analyze how efficiently the athlete is moving.
0:09:47 Are they generating energy and momentum in the direction that they want to generate in?
0:09:49 What is that desired direction?
0:09:54 So then we calculate all sorts of metrics around movement efficiency and direction.
0:09:58 And then once we calculate all those metrics, now we can understand how those relate to
0:10:00 what you’re trying to do.
0:10:02 Throw the ball as hard as possible.
0:10:07 So presumably with the picture, what you want to optimize for is having as much of the
0:10:14 picture’s energy of their body go toward making the ball go toward home plate.
0:10:15 Exactly.
0:10:16 Right?
0:10:17 I mean, is that that basic optimization problem?
0:10:18 Nailed it.
0:10:19 Yeah, yeah.
0:10:20 That’s exactly it.
0:10:21 And that’s the problem that we try to understand.
0:10:25 So when we build our sort of models regarding, like, how does a picture create efficient
0:10:30 fastball velocity, one of the most important things that comes out of those models is lining
0:10:34 up the direction of your torso rotation with the direction of your arm rotation.
0:10:36 The pitching motion is crazy complex, right?
0:10:41 Like they kick their leg up and they got their front arms doing something and their back arms
0:10:42 going back.
0:10:43 Yeah.
0:10:44 Like a lot is happening.
0:10:45 Yeah.
0:10:50 The sort of platonic ideal is every little millimeter of every motion is going toward maximizing
0:10:52 the energy of the ball going toward home plate.
0:10:53 Yes.
0:10:59 And not just maximizing like in a vacuum, but doing it in the most efficient way.
0:11:03 Because if you just sort of maximize in a vacuum, maybe you’re transferring that energy
0:11:07 in a way that hurts your elbow, you’re transferring that energy in a way that hurts your shoulder.
0:11:11 So not only do we figure out how can a picture maximize it, but we try to figure out how
0:11:17 they can have that energy and momentum go in a direction that doesn’t hurt their joints.
0:11:22 So try to have them throw the ball a little harder while also reducing their injury risk
0:11:23 a little bit.
0:11:33 So you’re whatever doing the math at Reboot HQ and then what are you sending back to the
0:11:35 team?
0:11:40 So some teams, so sort of again, all right, I told you that was the end of the biomechanics
0:11:41 train analogy.
0:11:44 I’m going to bring it back for a brief second.
0:11:45 Okay.
0:11:46 I’m ready.
0:11:50 So we go all the way to building a report that has a bunch of suggestions.
0:11:53 So there’s a report that’s like, this is how efficient you are.
0:11:56 You can like tilt your torso a little bit to be a little bit more efficient.
0:11:59 You can tilt your arm a little bit to be a little bit more efficient.
0:12:02 So we go all the way to generating a report.
0:12:04 That’s like the final stop on the train.
0:12:06 And that’s a report for one pitcher?
0:12:07 Yeah.
0:12:09 That’s a report on a pitcher for whatever.
0:12:15 And in a way that’s like the nerdiest, it’s what a pitching coach does, but just in a way
0:12:16 nerdy or way.
0:12:17 Yeah.
0:12:23 Well, so we try to sort of like give the coaches superpowers, you know, even though that, you
0:12:27 know, the coaches like can look at an athlete and understand a lot about the athlete just
0:12:29 by looking at them.
0:12:35 We try to make a report that can really amplify what the coach is already doing, maybe help
0:12:40 them discover some things they weren’t thinking about or measure some things that they weren’t
0:12:43 thinking about, but now they can track those a little bit easier.
0:12:47 So that’s, you know, that’s the ultimate thing that we produce is a report that can sort
0:12:51 of, like we said, give a coach superpowers.
0:12:53 Where was the train metaphor doing that?
0:12:56 Is there an earlier station of disembarkation?
0:12:57 Exactly.
0:12:58 Yeah.
0:13:04 So a lot of teams now are hiring people with biomechanics and analytics backgrounds.
0:13:09 So rather than just use our reports out of the box, they want to build their own reports
0:13:12 and their own statistical models and their own AI models.
0:13:16 So we also get, let people get off the train a little bit earlier and build whatever they
0:13:19 want on top of the data that we’re generating.
0:13:22 Are they not going to disintermediate you?
0:13:26 Once those people are there, does that reduce the value you provide to the team?
0:13:29 No, because we still have to process all that data.
0:13:35 Do you have some like IP or like, why can’t somebody like you who works for a team just
0:13:36 do that without you?
0:13:38 That is a great question.
0:13:45 We answer this question all the time is because it’s a very complex engineering problem not
0:13:49 only to do all the math, the physics-based math to calculate the energy, calculate the
0:13:56 momentum, like all of that math is really hard, but also to do it at a very large scale.
0:14:02 So someone like me in grad school learned how to do this on a sample size, you know,
0:14:07 my PhD was actually really just one pitch, but lots of people do it on like 10 pitches
0:14:09 or maybe 100 pitches.
0:14:18 So we do it on several thousand pitches like and swings like every morning, you know, there’s
0:14:23 just every team has like seven affiliates.
0:14:26 So there’s 150 games every day that need to be processed.
0:14:28 So you’re doing this on farm teams as well.
0:14:29 Yeah, exactly.
0:14:34 So not only did we solve the problem of doing it for like one pitcher in a way that’s really
0:14:40 actionable, but we solved the problem doing it for every game every day, you know, so
0:14:44 that you have the data when you wake up.
0:14:46 So you guys are doing it at scale.
0:14:51 The answer to why is that there is, in fact, an economy of scale, a benefit of scale and
0:14:52 that’s what you have.
0:14:53 Yes, exactly.
0:14:54 Yeah.
0:14:55 Yeah.
0:15:00 What’s a specific example of a thing that a coach might tell a pitcher in response to
0:15:04 your report to try and get them to throw whatever differently?
0:15:14 A really common low hanging fruit type of piece of feedback that often comes out of the reports
0:15:19 is how a pitcher is using their lead arm.
0:15:24 In a typical pitching motion, the pitcher will reach forward with their lead arm.
0:15:27 So just to be clear, the lead arm is the arm that is not holding the ball.
0:15:28 Right.
0:15:29 Yeah.
0:15:34 A pitcher will reach forward with that lead arm while the rear arm is holding the baseball
0:15:40 and they’ll rotate that lead arm really hard and that’s the thing that kind of initiates
0:15:41 the torso rotation.
0:15:49 So a very common flaw that we see is if a pitcher has a very vertical pitching arm,
0:15:54 they’re pitching the ball way over the top of their head, but their lead arm, when they
0:15:59 pull it through, when they swing it through, they swing it in a very flat plane.
0:16:01 Like across horizontal.
0:16:02 Horizontal.
0:16:03 Right.
0:16:04 Yeah.
0:16:07 That is not a very efficient plane to use when you’re throwing the ball on a very vertical
0:16:08 plane.
0:16:15 So a very common low hanging piece of fruit feedback that comes out of the report is having
0:16:20 pitchers just try to rotate their lead arm, pull with their lead arm in a more vertical
0:16:24 plane to better match what their torso is doing and better match what their pitching
0:16:26 arm is doing.
0:16:27 That’s a good one.
0:16:31 It’s like that one’s so simple that you don’t want it to get out because everybody’ll just
0:16:32 start looking at it.
0:16:37 No, I mean really, I mean like this has happened when I went to talk to a team and we talked
0:16:41 about some pieces, you know, some low hanging fruit and they’re like, “Okay, great.
0:16:44 We’ll take the lead arm thing and we’ll implement it everywhere.”
0:16:48 And I’m like, “Well, what about like the other 10 pages of the report?”
0:16:52 They’re like, “Nah, the lead arm, we’re good with the lead arm thing.”
0:16:54 Thanks, bye.
0:17:00 So for you, the end of the train is the report, but that report goes to the coach, right?
0:17:03 And so presumably the meaningful change hasn’t happened yet, right?
0:17:06 It has to somehow get from the coach to the pitcher.
0:17:11 And like, I know that piece of it is not your business now, but it was kind of your business
0:17:13 when you were at the Dodgers.
0:17:14 Presumably you’re familiar with it now.
0:17:16 Like, how does that piece of it work?
0:17:19 Is it like the pitching coach is like reading from the report to the pitcher?
0:17:21 I imagine not, but I don’t know.
0:17:25 No, definitely, definitely not.
0:17:32 Even at the Dodgers, my role was not being the one to coach the players.
0:17:35 It’s like, whatever you do, don’t talk to the pitchers, man.
0:17:37 Go back to your computer.
0:17:38 No, no.
0:17:45 I mean, thankfully, I got to be in the room as, you know, while the interaction is happening.
0:17:53 But I think that is the art of coaching that is so important is understanding the pitcher
0:17:58 and how the pitcher thinks about themselves and giving the right feedback to have the
0:18:03 pitcher do the thing that you want them to do.
0:18:08 Knowing how to talk to a player in a way that is not generic, presumably.
0:18:12 Different pitchers need to hear different things, even if the outcome is the same.
0:18:13 Right.
0:18:18 And there’s a, there’s a classic like debate in baseball of like, do you swing down on
0:18:21 the ball or do you swing with an uppercut?
0:18:24 And in reality, like the bat path is an arc.
0:18:27 The path goes down and then the path goes up.
0:18:33 But some coach, some players respond better when a coach will say, swing down on the ball.
0:18:36 And some players respond better when a coach will say, you know, have, get a little bit
0:18:41 more uppercut to your swing, but in reality, you know, the bat goes down and the bat goes
0:18:42 up.
0:18:46 So understanding what is the player, what helps the player the most.
0:18:49 So I heard you use this phrase in another interview, but I think is kind of what you’re
0:18:50 talking about here.
0:18:53 And it’s feel versus real.
0:18:54 What is that?
0:18:55 What is feel versus real?
0:19:01 So it’s exactly what we’re talking about is sometimes what the athlete feels like they’re
0:19:05 doing is not what’s actually happening.
0:19:11 But if you understand what the athlete feels like they’re doing, you can give feedback that
0:19:13 interacts with how they’re feeling.
0:19:18 So if they feel like they’re swinging down on the ball and you give them feedback related
0:19:22 to that, even if they’re actually swinging with an uppercut, you know, the important thing
0:19:26 is like, how do they feel and how do you give them feedback related to how they feel, which
0:19:32 then impacts what is real, but it’s understanding the interplay between the feel and the real.
0:19:38 Yeah, it’s wild that like there is, there is a human being doing a thing and that there
0:19:45 is this huge industrial machinery of your company and the team and all of these scientists
0:19:50 and all these cameras and computers that are trying to get this human being to change their
0:19:52 behavior in a very subtle way.
0:19:53 Yeah, exactly.
0:19:57 It’s a behavior that is in many ways intuitive, right?
0:20:00 It’s sort of partly conscious, but partly intuitive.
0:20:06 Like there’s a really interesting human being at the center of all of this.
0:20:11 Yeah, and I think that’s also why it’s so important to have the coach in the loop, because
0:20:17 they understand the human being even beyond just like the feel versus real aspect of like
0:20:22 did the athlete not get a good night’s sleep last night, then maybe today is not a good
0:20:23 day to give them feedback.
0:20:27 Are they going through challenges, you know, with a significant other, are they going through
0:20:30 challenges in other ways, or today are they really fired up?
0:20:34 So today is a really good day to give them feedback is like understanding the human being.
0:20:37 I’ll give them a shout out.
0:20:42 Connor Reginas is the assistant pitching coach right now for the Dodgers, and he was so good
0:20:43 at this.
0:20:50 He would always talk about he never felt comfortable truly coaching a player until the player trusted
0:20:54 him, until he felt like he had a good enough relationship with the player, as a player
0:20:57 would feel comfortable taking his feedback.
0:21:02 Like start with the player as a human being and then kind of get to the baseball.
0:21:08 And this is why I think I think it’s going to be really, really difficult to have a product
0:21:11 that goes direct to a player.
0:21:15 We’ve dabbled with that, you know, and lots of companies have dabbled with film yourself
0:21:20 with your iPhone and you get some feedback, you know, on your movement.
0:21:25 But I think there’s so much subjective stuff that goes into what we’re talking about.
0:21:29 How does the athlete move and how do they feel like they’re moving that I think it’s
0:21:36 going to be really hard to solve the challenge of giving an athlete feedback without a coach?
0:21:42 In a minute, Jimmy’s work in the NBA and why figuring out how to help NBA players shoot
0:21:54 better is actually a really hard problem.
0:21:56 What are you doing in basketball?
0:22:01 Basketball is very, very cool because it’s a slightly different challenge.
0:22:07 In baseball, like we’ve been talking about, the challenge for a pitcher is mostly just
0:22:09 maximizing efficiency.
0:22:14 Or even for a hitter, you know, they’re reacting to the pitcher, but they’re still trying to
0:22:18 swing as hard as they can and hit the ball as far as they can.
0:22:21 So basketball is very different because you got a target.
0:22:25 You’re not trying to do this thing as hard as humanly possible.
0:22:27 The target is the hoop.
0:22:28 The target is to be clear.
0:22:29 The target is the hoop.
0:22:36 So it’s a really interesting kind of like motor control problem of no matter where you
0:22:43 are on the court, how good are you at getting the ball to do what you want it to do?
0:22:50 We’ve been finding that there seems to be lots of trade-offs that good shooters are making
0:22:55 regarding like, when do they release the ball in the course of their jump?
0:22:59 How high of an arc do they use when they release the ball?
0:23:02 How high do they have their release point when they release the ball?
0:23:08 So there’s all sorts of trade-offs that these shooters seem to be making related to how
0:23:13 good they are at controlling their own jump, controlling their own velocity, where they
0:23:16 are on the court, how close the nearest defender is.
0:23:18 It’s a very different problem.
0:23:19 Sounds way harder.
0:23:20 Sounds way, way harder.
0:23:22 Is that right?
0:23:27 Harder for you to sort of solve, to write a report that says, “Do this differently and
0:23:29 you’ll hit a higher percentage of your shots.”
0:23:30 Right.
0:23:31 Exactly.
0:23:32 Yeah.
0:23:33 Which makes it fun.
0:23:35 I love solving hard problems.
0:23:38 What have you solved in basketball so far?
0:23:43 The first challenge that we had to solve was the scale challenge in basketball.
0:23:48 There aren’t quite as many games in basketball.
0:23:55 But the data is a lot harder to parse because the events aren’t as distinct.
0:23:59 It’s not like, “This is a shot only for free throws.”
0:24:00 It’s a more continuous game.
0:24:01 It’s continuous motion.
0:24:06 So the first big challenge was how do we even just isolate the events that we care about?
0:24:08 Like when does a shot begin?
0:24:10 What is time equals zero for a shot?
0:24:12 The kind of debatable.
0:24:13 Right.
0:24:14 Exactly.
0:24:17 So that was the first big challenge that we had to solve, was just sort of like the
0:24:20 data engineering challenge.
0:24:29 And now, like I said, our reports are more than telling you how to be more or less efficient.
0:24:32 It’s trying to surface the trade-offs that you’re making.
0:24:34 Where in your jump are you releasing the ball?
0:24:36 How high is your release point?
0:24:38 What kind of arc are you using?
0:24:42 And how does that compare to other arcs you could be using?
0:24:44 What are the trade-offs?
0:24:51 A really, really interesting trade-off to me is how much arc are you putting on the
0:24:55 basketball, not just to like evade a defender.
0:25:02 But there’s a trade-off where if you shoot the ball at a higher arc, you have to use
0:25:06 more velocity to get the ball to go all the way to the rim.
0:25:07 Right.
0:25:10 Because it’s going to travel farther in total in space.
0:25:11 Right.
0:25:12 Yeah.
0:25:19 And if you are not the most coordinated human, it might be harder for you to add more velocity
0:25:22 in a really precise way.
0:25:27 So the more arc you have, the more prone you can be to what we would call like velocity
0:25:29 errors, overshooting, undershooting.
0:25:35 The advantage you get when you create more arc is if you imagine the ball coming down
0:25:40 from that arc and the angle with which it approaches the rim, it approaches at a steeper
0:25:44 angle, which means you literally have more rim to aim at.
0:25:50 So if you shoot higher, this is a Steph Curry thing, he has a really high arc.
0:25:54 Presumably, this is the hypothesis because he’s one of the most coordinated humans on
0:25:55 the planet.
0:25:56 Seems plausible.
0:25:57 Seems plausible.
0:26:04 So he can use a really high arc because he’s really good at controlling his velocity output.
0:26:09 So he can really dial in how hard he releases the ball, which means he gets the advantage
0:26:14 of having more rim to aim at, whereas somebody who’s bad at controlling their velocity output,
0:26:17 when they try to aim higher, they’ll just overshoot and undershoot.
0:26:23 So is the notion that this is something of an oversimplification, but that for any given
0:26:29 level of coordination, there is some optimal arc, and the more coordinated you are, the
0:26:33 higher the optimal arc would be for you, setting aside defense?
0:26:34 Maybe.
0:26:35 That’s the hypothesis.
0:26:36 Yeah.
0:26:39 But that seems like where what you were saying goes.
0:26:40 Exactly.
0:26:41 Yeah.
0:26:42 That’s the hypothesis.
0:26:43 So that’s what we’re trying to look into.
0:26:44 Okay.
0:26:47 I’ll be curious to see what you figure out.
0:26:49 Are free throws easier?
0:26:51 Did you think of starting with free throws?
0:26:52 Yeah.
0:26:59 Right now, honestly, the phase we’re at in basketball is mostly just collecting a lot
0:27:01 of data.
0:27:05 So a starting pitcher will throw 90 pitches in a game, and now you have a sample size
0:27:10 of 90 pitches that are all the same, whereas a basketball shooter only has a couple of
0:27:12 free throws in a game.
0:27:16 We’re working on ways with teams of collecting data in a practice setting.
0:27:17 Yeah.
0:27:23 So getting a lot of this data in a bigger chunks, but really at this point, it’s collecting
0:27:28 a lot of data so we can do some of this research to explore some of these hypotheses.
0:27:33 So it seems like in basketball, you’re where you were 10 years ago or something in baseball.
0:27:34 Right.
0:27:35 Right.
0:27:40 So basketball is sort of one kind of frontier.
0:27:44 It seems like one thing you’re trying to figure out and haven’t really cracked yet.
0:27:46 What are some of the other frontiers?
0:27:50 The other things you’re figuring out, whether it’s in baseball or in the fundamental technology
0:27:54 or whatever, what are you working on?
0:28:03 We want to try to have computer vision be even more accessible.
0:28:10 There’s been a lot and a lot of improvements over the last 10 years in computer vision
0:28:17 where you can do really good motion capture with just your iPhone, but still for certain
0:28:26 specialized movements, pitching, shooting, things like that, there’s still little ways
0:28:29 to go to get really, really good data straight from your iPhone.
0:28:35 So that’s one of the frontiers is continuing to try to help understand how to make computer
0:28:39 vision more accessible.
0:28:45 Another one is more fitness-based analysis.
0:28:49 It’s interesting to think about companies like Mirror and companies that have tried
0:28:57 to give people feedback in a fitness environment, but how do you give someone or how do you
0:29:01 give a strength and conditioning coach good feedback that can be used in a weight room
0:29:02 setting?
0:29:03 Yeah.
0:29:08 And then continuing to explore other emotions and other sports, like football is a really
0:29:14 interesting one because a lot of football is interacting with other humans.
0:29:15 Indeed.
0:29:16 Yeah.
0:29:23 I mean, that’s obvious, but how do you get a takeaway from two linemen interacting?
0:29:24 Yeah.
0:29:25 Things like that.
0:29:32 And I wonder, I mean, it seems like if you think of the line in football, they’re so
0:29:35 on top of each other, the defensive line and the defensive line, then I feel like vision
0:29:37 might not be what you want.
0:29:38 You might want sensors, right?
0:29:42 You might want pressure sensors in the linemen’s clothes or something.
0:29:43 I don’t know.
0:29:44 I’m just making that up.
0:29:46 But it seems like that might be more useful just because it’s hard to see what’s going
0:29:47 on in the line.
0:29:48 Yeah.
0:29:49 Yeah.
0:29:50 Yeah.
0:29:53 Except professional athletes don’t like wearing random things.
0:29:58 Well, aren’t people trying to make sensors woven into the clothes?
0:30:02 I mean, I feel like there are ways you could just be putting on your jersey or putting
0:30:03 on your pads or whatever.
0:30:04 Oh, yeah.
0:30:05 And they would have the sensors built in.
0:30:06 Yeah.
0:30:07 Yeah.
0:30:08 A hundred percent.
0:30:12 There’s lots of really cool technology of that is microscopic sensors that are just
0:30:16 woven into clothing, for sure.
0:30:21 So let’s talk for a second more about the consumer side.
0:30:23 You sort of touched on it and moved on.
0:30:27 I mean, are you actively working on that or is that just to like, yeah, it kind of seems
0:30:30 interesting, but not for now or too busy?
0:30:33 What’s happening on the consumer side?
0:30:37 We’re not actively tackling the consumer side right now.
0:30:42 And the reason why we started with professional sports is, one, it’s because it’s what I
0:30:43 know.
0:30:45 I worked for the Dodgers.
0:30:54 But also, because the value proposition of what we’re doing is so impactful, you try
0:30:59 to understand the relationship between people who have tried to put numbers on this.
0:31:03 People have estimated it’s in the millions of dollars, but the value of adding one mile
0:31:08 an hour of fastball velocity to a pitcher, people have valued that.
0:31:10 We have valued that in the millions of dollars.
0:31:17 Well, sure, you would value it in the millions of dollars, but what is the, I don’t know,
0:31:18 what’s a top pitcher make these days?
0:31:19 I don’t even know anymore.
0:31:25 I mean, you want to talk about Shohei Otani, half a billion.
0:31:26 Hundreds of millions, right?
0:31:29 Half a billion.
0:31:35 So right, so the marginal benefit has a very large value, like a million dollars against
0:31:36 a hundred million dollars.
0:31:40 One percent better if they’re making a hundred million dollars, that’s worth a million dollars,
0:31:41 presumably.
0:31:45 Yeah, we still have to get the pro sports teams to believe that.
0:31:51 Well, how many pro sports teams are paying you at this point about?
0:31:55 Close to 10 in Major League Baseball and a couple in the NBA.
0:31:57 The NBA is a lot newer.
0:31:58 And how big is the field?
0:32:03 Like what’s sort of the broader state of play in the field of, I don’t even know what to
0:32:04 say.
0:32:09 Biomechanics as a service seems like a niche construction of it, but what would you say?
0:32:10 Sports analytics?
0:32:13 I mean, I guess that’s not quite right.
0:32:16 How do you construct the broader field?
0:32:18 Sports analytics is definitely a part of it.
0:32:26 The data that we provide is novel or is different than traditional sports analytics.
0:32:31 So we don’t really have a lot of companies as competitors.
0:32:41 Honestly, our biggest competitors are teams wanting to try to do this type of thing internally.
0:32:45 Hire a bunch of data engineers, hire a bunch of software engineers, hire people with biomechanics
0:32:50 backgrounds and try to build out these processing pipelines themselves.
0:32:55 Does every team have somebody with a PhD in biomechanics working for them now?
0:32:56 In baseball?
0:32:57 Yeah.
0:32:58 Wow.
0:33:02 Basketball, not yet, but they’re starting to.
0:33:07 So I know people, I mean, I think in general, baseball fans like to complain.
0:33:12 But so one of the things they have complained about lately is the way analytics more broadly
0:33:14 made the game more boring, right?
0:33:20 Like the shift and changing pictures more frequently and whatever else people complain
0:33:21 about.
0:33:26 Are you, do you fit into that at all?
0:33:35 Oh, that’s a good question.
0:33:41 Some people complain about how hard pictures are throwing these days, because it creates
0:33:45 more strikeouts and people think strikeouts are boring.
0:33:47 So maybe, yeah.
0:33:49 So maybe you need to get better at helping hitters.
0:33:53 That way you can even it back up.
0:33:54 Right.
0:33:55 And that’s what we talked about.
0:33:59 Take aways for hitters are harder because they’re reacting to the picture.
0:34:05 So if you think about the field, what your company say in five years or whatever’s your
0:34:11 kind of medium term future that you think about, what is the company and the sort of
0:34:16 the world, the sports world that you’re interacting with look like at that time and say whatever,
0:34:18 five years, 10 years?
0:34:30 What I hope is that what we are trying to foster is an environment where coaches have
0:34:35 really incredible tools at their disposal to understand how an athlete moves.
0:34:40 So we talked about at the very beginning where sort of the still more or less the state of
0:34:45 the art is a coach just looks at an athlete, watches video of an athlete and tries to give
0:34:50 the athlete feedback regarding what they see on the video, what they see with their eyes.
0:34:59 We hope that the standard becomes you use an analytical tool to help you understand how
0:35:04 the athlete is moving and to really like level up your coaching because now you have this
0:35:08 objective information about how the athlete is moving.
0:35:13 I kind of make the analogy related to just radar guns.
0:35:18 Before radar guns were a thing, a coach would just like look at a picture and be like, “Oh,
0:35:26 that looks pretty fast, make an adjustment and I think that looks a little faster.”
0:35:29 But then radar guns came out and you could actually measure how fast the picture was
0:35:34 throwing and you could actually measure if the picture is getting throwing the ball harder
0:35:37 based on your feedback.
0:35:42 And now you’re doing that but in a way more complex way.
0:35:43 Exactly.
0:35:44 Yeah.
0:36:01 We’ll be back in a minute with the lightning round.
0:36:02 Let’s do the lightning round.
0:36:05 Let’s start with a few baseball questions.
0:36:09 Who’s the most underrated pitcher of all time?
0:36:12 Whoa.
0:36:16 Underrated pitcher of all time, oh my goodness.
0:36:22 I don’t know if I can give you one that’s all time because I don’t know if that’s fair,
0:36:23 I’m sure.
0:36:24 I’m not thinking of everybody.
0:36:25 Of your lifetime.
0:36:26 Of your lifetime.
0:36:29 I grew up a hardcore Red Sox fan.
0:36:32 I grew up in Rhode Island.
0:36:37 And the first one that comes to mind, mostly in the Red Sox atmosphere, but I wonder if
0:36:43 you could make a broader argument was Tim Wakefield, who was a knuckleball pitcher
0:36:44 for the Red Sox.
0:36:45 And what made…
0:36:46 Amazing physics.
0:36:49 Amazing physics and aerodynamics.
0:36:56 And the reason being is he just did so many things for the Red Sox, ate up so many innings
0:37:03 and was so effective closing, starting, whatever, but he never got incredible recognition because
0:37:07 it was a knuckleball that was going 55, 60 miles an hour.
0:37:12 What’s one thing you would change about baseball to make it more popular?
0:37:20 I do feel like actually the changes that are being made are good ones in reducing the amount
0:37:21 of downtime.
0:37:25 It’s what, a pitch clock is one?
0:37:26 Is that the way you’re thinking of?
0:37:27 Yeah.
0:37:28 Yeah, exactly.
0:37:29 Yeah.
0:37:30 A pitch clock.
0:37:33 The thing that’s challenging for me as a biomechanist is when you reduce the amount of time that
0:37:39 a pitcher has to throw, you theoretically could introduce more fatigue, which theoretically
0:37:41 also introduces more injury risk.
0:37:45 So this is something that we’ve been thinking about is like…
0:37:53 So for me, while I like that change that baseball is making to speed the game up, I think the
0:38:00 pitchers also need to train a little bit differently to be able to better withstand the shorter
0:38:03 rest time.
0:38:09 Are there aspects of your work and the changes you’ve seen over the course of your career
0:38:16 that illuminate sort of broader changes in computer vision and AI more generally?
0:38:17 Yeah.
0:38:22 In particular, the most important one has been the improvement in computer vision, because
0:38:29 computer vision at its core is artificial intelligence, neural networks, and as that
0:38:33 technology has gotten better and better, the latest and greatest, people always talk about
0:38:36 the transformer model really changed AI.
0:38:38 I mean, that changed computer vision too.
0:38:44 Lots of the modern, more modern computer vision models are based on transformers.
0:38:51 Which to be clear, the transformer model is what gives us chat GPT, the T in GPT is transformer.
0:38:54 So how has it affected the computer vision side?
0:38:58 Making the models more accurate and more efficient.
0:39:02 It used to be a couple of years ago when I tried to run a computer vision model on my
0:39:10 laptop, it could take an hour, just to analyze one pitch, one video, and now it takes a matter
0:39:13 of seconds.
0:39:20 And so what is the bigger implication of that beyond your work?
0:39:26 More and more data is available regarding how people move.
0:39:31 When I first started, it was really hard to have data in a baseball game.
0:39:36 Now every major league game, every minor league game, every NBA game, maybe every G-league
0:39:46 game, every WNBA game, every single basketball and baseball game, more or less, is now being
0:39:51 recorded with computer vision to get the three dimensional data about how people are moving.
0:39:57 Just lots and lots of data on how people move, and this is really impacting, I think, lots
0:40:04 of fields, in particular, I think, like self-driving cars, robots that are meant to interact with
0:40:07 the world, all rely on computer vision models.
0:40:12 I mean, I think one of the coolest things about how cars do this sort of thing is not
0:40:21 only do they have to understand where a person is now, but they’re really cool models that
0:40:28 they take where a person is now and the last 10 seconds of what that person did and try
0:40:31 to predict all the different things that the person might do.
0:40:36 They might run across the street, they might jump out of the way, they might jump forward,
0:40:38 they might run and chase a soccer ball.
0:40:44 I mean, my sense is all of the hard edge cases in self-driving cars, basically the reason
0:40:50 we don’t truly have self-driving cars yet, is because people are so hard to understand.
0:40:55 If the world was all self-driving cars, then it would be a solved problem, right?
0:40:59 The machine could understand what other machines are going to do, but people, human drivers,
0:41:08 human pedestrians are strange and very hard for machines to understand at the same time.
0:41:09 It’s like they need a coach.
0:41:13 It’s like the coaching problem.
0:41:18 We found your sound cloud.
0:41:21 So Jimmy Buffett fans are called parrot heads.
0:41:26 What are Jimmy Buffett fans called?
0:41:31 I’ve never thought about it.
0:41:38 You have a song called “Let’s Have Some Fruit”, parenthesis, the fruit song.
0:41:41 Is fruit a metaphor?
0:41:43 Leave it up to your imagination.
0:41:45 Fair.
0:41:50 Jimmy Buffett is the co-founder and CEO of Reboot Motion.
0:42:06 Today’s show was produced by Gabriel Hunter Chang.
0:42:11 It was edited by Lydia Jean Kott and engineered by Sarah Brugger.
0:42:14 You can email us at problem@pushkin.fm.
0:42:18 I’m Jacob Goldstein and we’ll be back next week with another episode of What’s Your Problem.
0:42:19 I’m Jacob Goldstein and we’ll be back next week with another episode of What’s Your Problem.
0:42:21 (upbeat music)
0:42:23 (upbeat music)
0:00:24 school and initially start working on prosthetic limbs.
0:00:25 Okay.
0:00:29 So that was the first two years of grad school.
0:00:35 Then about two years in, I discovered baseball pitching biomechanics research.
0:00:40 The funny thing that happened there was I gave a Ph.D. committee meeting where I spent
0:00:45 most of the time talking about prosthetic limbs and then I spent the last few minutes
0:00:51 as an aside on the baseball pitching research I found and my committee was like, “Jimmy,
0:00:55 the last few minutes were way better than the first 45.”
0:00:57 I was like, “All right.”
0:01:02 So then they were like, “We’ll let you do baseball pitching biomechanics as your Ph.D.
0:01:08 work, but just so you know, it might be really hard to have a career doing that.”
0:01:10 But they were like, “If you want to go for it, you can go for it.”
0:01:19 And so I was like, “You know what, sure, I’ll go for it.”
0:01:22 I’m Jacob Goldstein and this is What’s Your Problem.
0:01:27 Today we have the second episode in our series about people who are working at the frontiers
0:01:31 of technology to help elite athletes perform better.
0:01:34 My guest today is Jimmy Buffy.
0:01:39 And as it happened, the concerns of his grad school advisors were unfounded.
0:01:44 Jimmy has in fact made a career out of the biomechanics of pitching in baseball and sports
0:01:47 biomechanics more broadly.
0:01:51 When he finished grad school, he got a job with the Los Angeles Dodgers and he went on
0:01:56 to co-found a company called Reboot Motion that works with teams in Major League Baseball
0:01:57 and the NBA.
0:01:59 Jimmy’s problem is this.
0:02:05 How do you take massive amounts of data about how professional athletes move and turn all
0:02:12 that data into information that actually helps those athletes perform better?
0:02:17 You end up doing your dissertation research on the biomechanics of pitching, of baseball,
0:02:20 pitching in baseball.
0:02:23 And then you get hired by the Dodgers.
0:02:24 Yeah.
0:02:25 Yeah.
0:02:26 That was awesome.
0:02:29 Because I originally didn’t, again, didn’t realize that that could be a thing that could
0:02:30 happen.
0:02:32 Well, and it kind of wasn’t, right?
0:02:37 Like you’re kind of just coming into this field as it’s becoming a field where you can
0:02:39 get a job, where it’s a field, right?
0:02:40 Right, exactly.
0:02:41 Yeah.
0:02:44 I mean, the challenge then was there wasn’t a lot of options for actually even getting
0:02:47 the data that you need to analyze.
0:02:53 So 10 years ago, like what is the state of play in this sort of nascent field that you’re
0:02:56 in helping to create?
0:03:00 So the field, I would say, is like sports biomechanics.
0:03:06 And what that is, is being able to analyze the movement of athletes for lots of purposes,
0:03:10 help them reduce injury risk, help them improve performance.
0:03:15 And to be clear, like folk sports biomechanics has been around forever, right?
0:03:16 That’s what coaches do.
0:03:17 They stand there.
0:03:18 Yeah.
0:03:21 And so it’s kind of becoming, it’s becoming more technical, right?
0:03:22 The field is becoming more technical.
0:03:23 Yeah.
0:03:31 So the state of the art relied on what is called marker-based motion capture, which is where
0:03:36 you literally put reflective markers, like little balls, you stick them all over somebody’s
0:03:37 body.
0:03:41 Usually, the person has to like strip their clothes off because you want the markers like
0:03:47 literally like on the skin, on the joints, and then you have these special cameras that
0:03:48 track those markers.
0:03:53 So you’ve got like a pitcher in his underwear with a bunch of little metal balls, and they’re
0:03:55 like, just pitch like you always pitch.
0:03:56 Right.
0:03:57 And that’s the challenge.
0:04:02 That’s why it wasn’t very widespread as a thing people did because it was so hard to
0:04:07 collect the data because ultimately what you would need is you need the data on how someone
0:04:08 is moving.
0:04:12 You need to track where their elbow is, where their wrist is, where their knees are so that
0:04:13 you can analyze it.
0:04:17 And the state of the art for tracking it was an awful experience for the people you were
0:04:18 trying to track.
0:04:22 And that presumably would mean they didn’t pitch like they usually pitch.
0:04:23 Exactly.
0:04:25 Because they don’t usually stand there in their underwear with metal.
0:04:26 Were they really in their underwear?
0:04:29 By the way, I’m saying it because it sounds absurd, but is that actually what they were
0:04:30 doing?
0:04:34 That’s actually, you strip down to your boxers, your boxer briefs, and that’s it.
0:04:35 That’s all you’re wearing.
0:04:39 So it basically didn’t work and it basically wasn’t very widely used as a result.
0:04:40 Right.
0:04:41 Exactly.
0:04:47 Yeah, you look at studies in that field and people would be throwing like several miles
0:04:52 an hour slower than they would be throwing when they weren’t wearing all that stuff.
0:04:54 So what changes?
0:04:56 Computer vision.
0:04:59 That was the big inflection point.
0:05:06 Now to be fair, even when I was finishing my PhD, and I’ll give them a shout out, there
0:05:12 was a company that was already working really hard to solve this problem for baseball teams
0:05:18 that I got to be familiar with called Kinatracks.
0:05:23 But yeah, the big inflection point was computer vision, basically using artificial intelligence
0:05:31 to identify where those joints are in a camera image rather than needing to paste those reflective
0:05:32 markers.
0:05:38 So computer vision takes off, you’re working at the Dodgers, and then eventually in 2019,
0:05:43 you leave the Dodgers and you start your company reboot motion.
0:05:45 What does your company do?
0:05:49 We do what we call biomechanics as a service.
0:05:55 So we try to analyze this computer vision data at a very large scale to help teams and
0:05:58 coaches make use of it to help athletes get better.
0:05:59 Bass?
0:06:00 Bass, yeah.
0:06:05 That’s bass, but bass.
0:06:07 And who are your customers?
0:06:16 Our customers are majorly baseball teams, actually NBA teams, so we’ve gotten into basketball.
0:06:21 So sort of like league wide data providers.
0:06:24 So yeah, leagues, teams is basically our sweet spot.
0:06:30 So let’s talk in like a little more detail about what you actually do, right?
0:06:33 Tell me the story of what you do.
0:06:39 So Evan, my co-founder, Evan Demchick, he likes to call this the biomechanics train.
0:06:43 Hey, let’s take a ride on the biomechanics train.
0:06:49 And we call it that because the way our product works is we let people get on the train at
0:06:53 whatever stop works for them and get off the train at whatever stop works for them.
0:06:55 How far are we going to go with this metaphor?
0:06:57 I’m a little, I’m nervous now about it.
0:06:59 That might be as far as we go.
0:07:00 Okay.
0:07:01 Good.
0:07:02 Good.
0:07:09 So just tell it to me, start at whatever seems like the beginning of an encounter.
0:07:10 And so I’d like to understand how that works.
0:07:13 That’s really kind of a way to think about it.
0:07:16 So let’s start with the data.
0:07:19 What is the basic thing that’s happening?
0:07:27 So the very first thing that happens is you record videos of the athlete doing the athletic
0:07:28 motion.
0:07:29 So we’ll talk about pitching.
0:07:32 So you record videos of a pitcher pitching.
0:07:35 Our product, we actually have implemented our own computer vision models.
0:07:38 So we can do that if people want.
0:07:44 But generally speaking, people have systems like kinetraxx is one, I’ve mentioned that
0:07:49 Hawkeye is another popular one where there is a system in place that has the cameras
0:07:53 that record the videos and then runs those computer vision models.
0:07:58 And what those computer vision models do is they extract the locations in three dimensional
0:08:01 space of all of like the joint centers.
0:08:02 So where’s my elbow?
0:08:03 Where’s my wrist?
0:08:04 Where’s my knee?
0:08:06 In three dimensional space.
0:08:09 That’s what comes out of these computer vision systems.
0:08:12 So pretty much everybody has that at this point?
0:08:17 Like every professional baseball team has that for every pitch in every game at this
0:08:18 point?
0:08:19 Yes, exactly.
0:08:21 So it’s a ton, a ton of data.
0:08:25 So that’s another challenge that we’ve solved is not only how do you do this, but how do
0:08:28 you do this at a very large scale?
0:08:30 So this data, everybody’s got it.
0:08:35 And you’re not, you can sort of process the data, but that’s not your special sauce.
0:08:36 That’s not your secret sauce, right?
0:08:40 So typically they’ll send you that data of like, here’s all the body points, here’s how
0:08:43 they’re moving in physical space.
0:08:44 Then what do you do with it?
0:08:47 So yeah, this is where the special sauce comes in.
0:08:52 So the first step to that is how do you turn those key points into a human skeleton?
0:08:55 So you got to figure out like, where do the bones connect?
0:08:58 What sort of degrees of freedom do those bones have?
0:09:05 So then you can figure out how do those key points animate a human skeleton?
0:09:06 So you’re sort of rebuilding it.
0:09:10 It’s like you start with a picture of a person and then you turn it into a bunch of data
0:09:11 points.
0:09:13 And then you got to kind of build the person back up again from the database.
0:09:14 Exactly, exactly.
0:09:20 So then, so now you have an actual human skeleton where the shoulder is rotating, the elbow
0:09:24 is flexing, the knee is flexing, the hips are rotating.
0:09:28 And now once you do that, now you can understand that data in the context of how the body
0:09:29 works.
0:09:30 Okay.
0:09:35 So once we’ve done that, then we calculate how energy flows through the body.
0:09:37 We calculate how momentum flows through the body.
0:09:41 And once we’ve done that, we can analyze how efficiently the athlete is moving.
0:09:47 Are they generating energy and momentum in the direction that they want to generate in?
0:09:49 What is that desired direction?
0:09:54 So then we calculate all sorts of metrics around movement efficiency and direction.
0:09:58 And then once we calculate all those metrics, now we can understand how those relate to
0:10:00 what you’re trying to do.
0:10:02 Throw the ball as hard as possible.
0:10:07 So presumably with the picture, what you want to optimize for is having as much of the
0:10:14 picture’s energy of their body go toward making the ball go toward home plate.
0:10:15 Exactly.
0:10:16 Right?
0:10:17 I mean, is that that basic optimization problem?
0:10:18 Nailed it.
0:10:19 Yeah, yeah.
0:10:20 That’s exactly it.
0:10:21 And that’s the problem that we try to understand.
0:10:25 So when we build our sort of models regarding, like, how does a picture create efficient
0:10:30 fastball velocity, one of the most important things that comes out of those models is lining
0:10:34 up the direction of your torso rotation with the direction of your arm rotation.
0:10:36 The pitching motion is crazy complex, right?
0:10:41 Like they kick their leg up and they got their front arms doing something and their back arms
0:10:42 going back.
0:10:43 Yeah.
0:10:44 Like a lot is happening.
0:10:45 Yeah.
0:10:50 The sort of platonic ideal is every little millimeter of every motion is going toward maximizing
0:10:52 the energy of the ball going toward home plate.
0:10:53 Yes.
0:10:59 And not just maximizing like in a vacuum, but doing it in the most efficient way.
0:11:03 Because if you just sort of maximize in a vacuum, maybe you’re transferring that energy
0:11:07 in a way that hurts your elbow, you’re transferring that energy in a way that hurts your shoulder.
0:11:11 So not only do we figure out how can a picture maximize it, but we try to figure out how
0:11:17 they can have that energy and momentum go in a direction that doesn’t hurt their joints.
0:11:22 So try to have them throw the ball a little harder while also reducing their injury risk
0:11:23 a little bit.
0:11:33 So you’re whatever doing the math at Reboot HQ and then what are you sending back to the
0:11:35 team?
0:11:40 So some teams, so sort of again, all right, I told you that was the end of the biomechanics
0:11:41 train analogy.
0:11:44 I’m going to bring it back for a brief second.
0:11:45 Okay.
0:11:46 I’m ready.
0:11:50 So we go all the way to building a report that has a bunch of suggestions.
0:11:53 So there’s a report that’s like, this is how efficient you are.
0:11:56 You can like tilt your torso a little bit to be a little bit more efficient.
0:11:59 You can tilt your arm a little bit to be a little bit more efficient.
0:12:02 So we go all the way to generating a report.
0:12:04 That’s like the final stop on the train.
0:12:06 And that’s a report for one pitcher?
0:12:07 Yeah.
0:12:09 That’s a report on a pitcher for whatever.
0:12:15 And in a way that’s like the nerdiest, it’s what a pitching coach does, but just in a way
0:12:16 nerdy or way.
0:12:17 Yeah.
0:12:23 Well, so we try to sort of like give the coaches superpowers, you know, even though that, you
0:12:27 know, the coaches like can look at an athlete and understand a lot about the athlete just
0:12:29 by looking at them.
0:12:35 We try to make a report that can really amplify what the coach is already doing, maybe help
0:12:40 them discover some things they weren’t thinking about or measure some things that they weren’t
0:12:43 thinking about, but now they can track those a little bit easier.
0:12:47 So that’s, you know, that’s the ultimate thing that we produce is a report that can sort
0:12:51 of, like we said, give a coach superpowers.
0:12:53 Where was the train metaphor doing that?
0:12:56 Is there an earlier station of disembarkation?
0:12:57 Exactly.
0:12:58 Yeah.
0:13:04 So a lot of teams now are hiring people with biomechanics and analytics backgrounds.
0:13:09 So rather than just use our reports out of the box, they want to build their own reports
0:13:12 and their own statistical models and their own AI models.
0:13:16 So we also get, let people get off the train a little bit earlier and build whatever they
0:13:19 want on top of the data that we’re generating.
0:13:22 Are they not going to disintermediate you?
0:13:26 Once those people are there, does that reduce the value you provide to the team?
0:13:29 No, because we still have to process all that data.
0:13:35 Do you have some like IP or like, why can’t somebody like you who works for a team just
0:13:36 do that without you?
0:13:38 That is a great question.
0:13:45 We answer this question all the time is because it’s a very complex engineering problem not
0:13:49 only to do all the math, the physics-based math to calculate the energy, calculate the
0:13:56 momentum, like all of that math is really hard, but also to do it at a very large scale.
0:14:02 So someone like me in grad school learned how to do this on a sample size, you know,
0:14:07 my PhD was actually really just one pitch, but lots of people do it on like 10 pitches
0:14:09 or maybe 100 pitches.
0:14:18 So we do it on several thousand pitches like and swings like every morning, you know, there’s
0:14:23 just every team has like seven affiliates.
0:14:26 So there’s 150 games every day that need to be processed.
0:14:28 So you’re doing this on farm teams as well.
0:14:29 Yeah, exactly.
0:14:34 So not only did we solve the problem of doing it for like one pitcher in a way that’s really
0:14:40 actionable, but we solved the problem doing it for every game every day, you know, so
0:14:44 that you have the data when you wake up.
0:14:46 So you guys are doing it at scale.
0:14:51 The answer to why is that there is, in fact, an economy of scale, a benefit of scale and
0:14:52 that’s what you have.
0:14:53 Yes, exactly.
0:14:54 Yeah.
0:14:55 Yeah.
0:15:00 What’s a specific example of a thing that a coach might tell a pitcher in response to
0:15:04 your report to try and get them to throw whatever differently?
0:15:14 A really common low hanging fruit type of piece of feedback that often comes out of the reports
0:15:19 is how a pitcher is using their lead arm.
0:15:24 In a typical pitching motion, the pitcher will reach forward with their lead arm.
0:15:27 So just to be clear, the lead arm is the arm that is not holding the ball.
0:15:28 Right.
0:15:29 Yeah.
0:15:34 A pitcher will reach forward with that lead arm while the rear arm is holding the baseball
0:15:40 and they’ll rotate that lead arm really hard and that’s the thing that kind of initiates
0:15:41 the torso rotation.
0:15:49 So a very common flaw that we see is if a pitcher has a very vertical pitching arm,
0:15:54 they’re pitching the ball way over the top of their head, but their lead arm, when they
0:15:59 pull it through, when they swing it through, they swing it in a very flat plane.
0:16:01 Like across horizontal.
0:16:02 Horizontal.
0:16:03 Right.
0:16:04 Yeah.
0:16:07 That is not a very efficient plane to use when you’re throwing the ball on a very vertical
0:16:08 plane.
0:16:15 So a very common low hanging piece of fruit feedback that comes out of the report is having
0:16:20 pitchers just try to rotate their lead arm, pull with their lead arm in a more vertical
0:16:24 plane to better match what their torso is doing and better match what their pitching
0:16:26 arm is doing.
0:16:27 That’s a good one.
0:16:31 It’s like that one’s so simple that you don’t want it to get out because everybody’ll just
0:16:32 start looking at it.
0:16:37 No, I mean really, I mean like this has happened when I went to talk to a team and we talked
0:16:41 about some pieces, you know, some low hanging fruit and they’re like, “Okay, great.
0:16:44 We’ll take the lead arm thing and we’ll implement it everywhere.”
0:16:48 And I’m like, “Well, what about like the other 10 pages of the report?”
0:16:52 They’re like, “Nah, the lead arm, we’re good with the lead arm thing.”
0:16:54 Thanks, bye.
0:17:00 So for you, the end of the train is the report, but that report goes to the coach, right?
0:17:03 And so presumably the meaningful change hasn’t happened yet, right?
0:17:06 It has to somehow get from the coach to the pitcher.
0:17:11 And like, I know that piece of it is not your business now, but it was kind of your business
0:17:13 when you were at the Dodgers.
0:17:14 Presumably you’re familiar with it now.
0:17:16 Like, how does that piece of it work?
0:17:19 Is it like the pitching coach is like reading from the report to the pitcher?
0:17:21 I imagine not, but I don’t know.
0:17:25 No, definitely, definitely not.
0:17:32 Even at the Dodgers, my role was not being the one to coach the players.
0:17:35 It’s like, whatever you do, don’t talk to the pitchers, man.
0:17:37 Go back to your computer.
0:17:38 No, no.
0:17:45 I mean, thankfully, I got to be in the room as, you know, while the interaction is happening.
0:17:53 But I think that is the art of coaching that is so important is understanding the pitcher
0:17:58 and how the pitcher thinks about themselves and giving the right feedback to have the
0:18:03 pitcher do the thing that you want them to do.
0:18:08 Knowing how to talk to a player in a way that is not generic, presumably.
0:18:12 Different pitchers need to hear different things, even if the outcome is the same.
0:18:13 Right.
0:18:18 And there’s a, there’s a classic like debate in baseball of like, do you swing down on
0:18:21 the ball or do you swing with an uppercut?
0:18:24 And in reality, like the bat path is an arc.
0:18:27 The path goes down and then the path goes up.
0:18:33 But some coach, some players respond better when a coach will say, swing down on the ball.
0:18:36 And some players respond better when a coach will say, you know, have, get a little bit
0:18:41 more uppercut to your swing, but in reality, you know, the bat goes down and the bat goes
0:18:42 up.
0:18:46 So understanding what is the player, what helps the player the most.
0:18:49 So I heard you use this phrase in another interview, but I think is kind of what you’re
0:18:50 talking about here.
0:18:53 And it’s feel versus real.
0:18:54 What is that?
0:18:55 What is feel versus real?
0:19:01 So it’s exactly what we’re talking about is sometimes what the athlete feels like they’re
0:19:05 doing is not what’s actually happening.
0:19:11 But if you understand what the athlete feels like they’re doing, you can give feedback that
0:19:13 interacts with how they’re feeling.
0:19:18 So if they feel like they’re swinging down on the ball and you give them feedback related
0:19:22 to that, even if they’re actually swinging with an uppercut, you know, the important thing
0:19:26 is like, how do they feel and how do you give them feedback related to how they feel, which
0:19:32 then impacts what is real, but it’s understanding the interplay between the feel and the real.
0:19:38 Yeah, it’s wild that like there is, there is a human being doing a thing and that there
0:19:45 is this huge industrial machinery of your company and the team and all of these scientists
0:19:50 and all these cameras and computers that are trying to get this human being to change their
0:19:52 behavior in a very subtle way.
0:19:53 Yeah, exactly.
0:19:57 It’s a behavior that is in many ways intuitive, right?
0:20:00 It’s sort of partly conscious, but partly intuitive.
0:20:06 Like there’s a really interesting human being at the center of all of this.
0:20:11 Yeah, and I think that’s also why it’s so important to have the coach in the loop, because
0:20:17 they understand the human being even beyond just like the feel versus real aspect of like
0:20:22 did the athlete not get a good night’s sleep last night, then maybe today is not a good
0:20:23 day to give them feedback.
0:20:27 Are they going through challenges, you know, with a significant other, are they going through
0:20:30 challenges in other ways, or today are they really fired up?
0:20:34 So today is a really good day to give them feedback is like understanding the human being.
0:20:37 I’ll give them a shout out.
0:20:42 Connor Reginas is the assistant pitching coach right now for the Dodgers, and he was so good
0:20:43 at this.
0:20:50 He would always talk about he never felt comfortable truly coaching a player until the player trusted
0:20:54 him, until he felt like he had a good enough relationship with the player, as a player
0:20:57 would feel comfortable taking his feedback.
0:21:02 Like start with the player as a human being and then kind of get to the baseball.
0:21:08 And this is why I think I think it’s going to be really, really difficult to have a product
0:21:11 that goes direct to a player.
0:21:15 We’ve dabbled with that, you know, and lots of companies have dabbled with film yourself
0:21:20 with your iPhone and you get some feedback, you know, on your movement.
0:21:25 But I think there’s so much subjective stuff that goes into what we’re talking about.
0:21:29 How does the athlete move and how do they feel like they’re moving that I think it’s
0:21:36 going to be really hard to solve the challenge of giving an athlete feedback without a coach?
0:21:42 In a minute, Jimmy’s work in the NBA and why figuring out how to help NBA players shoot
0:21:54 better is actually a really hard problem.
0:21:56 What are you doing in basketball?
0:22:01 Basketball is very, very cool because it’s a slightly different challenge.
0:22:07 In baseball, like we’ve been talking about, the challenge for a pitcher is mostly just
0:22:09 maximizing efficiency.
0:22:14 Or even for a hitter, you know, they’re reacting to the pitcher, but they’re still trying to
0:22:18 swing as hard as they can and hit the ball as far as they can.
0:22:21 So basketball is very different because you got a target.
0:22:25 You’re not trying to do this thing as hard as humanly possible.
0:22:27 The target is the hoop.
0:22:28 The target is to be clear.
0:22:29 The target is the hoop.
0:22:36 So it’s a really interesting kind of like motor control problem of no matter where you
0:22:43 are on the court, how good are you at getting the ball to do what you want it to do?
0:22:50 We’ve been finding that there seems to be lots of trade-offs that good shooters are making
0:22:55 regarding like, when do they release the ball in the course of their jump?
0:22:59 How high of an arc do they use when they release the ball?
0:23:02 How high do they have their release point when they release the ball?
0:23:08 So there’s all sorts of trade-offs that these shooters seem to be making related to how
0:23:13 good they are at controlling their own jump, controlling their own velocity, where they
0:23:16 are on the court, how close the nearest defender is.
0:23:18 It’s a very different problem.
0:23:19 Sounds way harder.
0:23:20 Sounds way, way harder.
0:23:22 Is that right?
0:23:27 Harder for you to sort of solve, to write a report that says, “Do this differently and
0:23:29 you’ll hit a higher percentage of your shots.”
0:23:30 Right.
0:23:31 Exactly.
0:23:32 Yeah.
0:23:33 Which makes it fun.
0:23:35 I love solving hard problems.
0:23:38 What have you solved in basketball so far?
0:23:43 The first challenge that we had to solve was the scale challenge in basketball.
0:23:48 There aren’t quite as many games in basketball.
0:23:55 But the data is a lot harder to parse because the events aren’t as distinct.
0:23:59 It’s not like, “This is a shot only for free throws.”
0:24:00 It’s a more continuous game.
0:24:01 It’s continuous motion.
0:24:06 So the first big challenge was how do we even just isolate the events that we care about?
0:24:08 Like when does a shot begin?
0:24:10 What is time equals zero for a shot?
0:24:12 The kind of debatable.
0:24:13 Right.
0:24:14 Exactly.
0:24:17 So that was the first big challenge that we had to solve, was just sort of like the
0:24:20 data engineering challenge.
0:24:29 And now, like I said, our reports are more than telling you how to be more or less efficient.
0:24:32 It’s trying to surface the trade-offs that you’re making.
0:24:34 Where in your jump are you releasing the ball?
0:24:36 How high is your release point?
0:24:38 What kind of arc are you using?
0:24:42 And how does that compare to other arcs you could be using?
0:24:44 What are the trade-offs?
0:24:51 A really, really interesting trade-off to me is how much arc are you putting on the
0:24:55 basketball, not just to like evade a defender.
0:25:02 But there’s a trade-off where if you shoot the ball at a higher arc, you have to use
0:25:06 more velocity to get the ball to go all the way to the rim.
0:25:07 Right.
0:25:10 Because it’s going to travel farther in total in space.
0:25:11 Right.
0:25:12 Yeah.
0:25:19 And if you are not the most coordinated human, it might be harder for you to add more velocity
0:25:22 in a really precise way.
0:25:27 So the more arc you have, the more prone you can be to what we would call like velocity
0:25:29 errors, overshooting, undershooting.
0:25:35 The advantage you get when you create more arc is if you imagine the ball coming down
0:25:40 from that arc and the angle with which it approaches the rim, it approaches at a steeper
0:25:44 angle, which means you literally have more rim to aim at.
0:25:50 So if you shoot higher, this is a Steph Curry thing, he has a really high arc.
0:25:54 Presumably, this is the hypothesis because he’s one of the most coordinated humans on
0:25:55 the planet.
0:25:56 Seems plausible.
0:25:57 Seems plausible.
0:26:04 So he can use a really high arc because he’s really good at controlling his velocity output.
0:26:09 So he can really dial in how hard he releases the ball, which means he gets the advantage
0:26:14 of having more rim to aim at, whereas somebody who’s bad at controlling their velocity output,
0:26:17 when they try to aim higher, they’ll just overshoot and undershoot.
0:26:23 So is the notion that this is something of an oversimplification, but that for any given
0:26:29 level of coordination, there is some optimal arc, and the more coordinated you are, the
0:26:33 higher the optimal arc would be for you, setting aside defense?
0:26:34 Maybe.
0:26:35 That’s the hypothesis.
0:26:36 Yeah.
0:26:39 But that seems like where what you were saying goes.
0:26:40 Exactly.
0:26:41 Yeah.
0:26:42 That’s the hypothesis.
0:26:43 So that’s what we’re trying to look into.
0:26:44 Okay.
0:26:47 I’ll be curious to see what you figure out.
0:26:49 Are free throws easier?
0:26:51 Did you think of starting with free throws?
0:26:52 Yeah.
0:26:59 Right now, honestly, the phase we’re at in basketball is mostly just collecting a lot
0:27:01 of data.
0:27:05 So a starting pitcher will throw 90 pitches in a game, and now you have a sample size
0:27:10 of 90 pitches that are all the same, whereas a basketball shooter only has a couple of
0:27:12 free throws in a game.
0:27:16 We’re working on ways with teams of collecting data in a practice setting.
0:27:17 Yeah.
0:27:23 So getting a lot of this data in a bigger chunks, but really at this point, it’s collecting
0:27:28 a lot of data so we can do some of this research to explore some of these hypotheses.
0:27:33 So it seems like in basketball, you’re where you were 10 years ago or something in baseball.
0:27:34 Right.
0:27:35 Right.
0:27:40 So basketball is sort of one kind of frontier.
0:27:44 It seems like one thing you’re trying to figure out and haven’t really cracked yet.
0:27:46 What are some of the other frontiers?
0:27:50 The other things you’re figuring out, whether it’s in baseball or in the fundamental technology
0:27:54 or whatever, what are you working on?
0:28:03 We want to try to have computer vision be even more accessible.
0:28:10 There’s been a lot and a lot of improvements over the last 10 years in computer vision
0:28:17 where you can do really good motion capture with just your iPhone, but still for certain
0:28:26 specialized movements, pitching, shooting, things like that, there’s still little ways
0:28:29 to go to get really, really good data straight from your iPhone.
0:28:35 So that’s one of the frontiers is continuing to try to help understand how to make computer
0:28:39 vision more accessible.
0:28:45 Another one is more fitness-based analysis.
0:28:49 It’s interesting to think about companies like Mirror and companies that have tried
0:28:57 to give people feedback in a fitness environment, but how do you give someone or how do you
0:29:01 give a strength and conditioning coach good feedback that can be used in a weight room
0:29:02 setting?
0:29:03 Yeah.
0:29:08 And then continuing to explore other emotions and other sports, like football is a really
0:29:14 interesting one because a lot of football is interacting with other humans.
0:29:15 Indeed.
0:29:16 Yeah.
0:29:23 I mean, that’s obvious, but how do you get a takeaway from two linemen interacting?
0:29:24 Yeah.
0:29:25 Things like that.
0:29:32 And I wonder, I mean, it seems like if you think of the line in football, they’re so
0:29:35 on top of each other, the defensive line and the defensive line, then I feel like vision
0:29:37 might not be what you want.
0:29:38 You might want sensors, right?
0:29:42 You might want pressure sensors in the linemen’s clothes or something.
0:29:43 I don’t know.
0:29:44 I’m just making that up.
0:29:46 But it seems like that might be more useful just because it’s hard to see what’s going
0:29:47 on in the line.
0:29:48 Yeah.
0:29:49 Yeah.
0:29:50 Yeah.
0:29:53 Except professional athletes don’t like wearing random things.
0:29:58 Well, aren’t people trying to make sensors woven into the clothes?
0:30:02 I mean, I feel like there are ways you could just be putting on your jersey or putting
0:30:03 on your pads or whatever.
0:30:04 Oh, yeah.
0:30:05 And they would have the sensors built in.
0:30:06 Yeah.
0:30:07 Yeah.
0:30:08 A hundred percent.
0:30:12 There’s lots of really cool technology of that is microscopic sensors that are just
0:30:16 woven into clothing, for sure.
0:30:21 So let’s talk for a second more about the consumer side.
0:30:23 You sort of touched on it and moved on.
0:30:27 I mean, are you actively working on that or is that just to like, yeah, it kind of seems
0:30:30 interesting, but not for now or too busy?
0:30:33 What’s happening on the consumer side?
0:30:37 We’re not actively tackling the consumer side right now.
0:30:42 And the reason why we started with professional sports is, one, it’s because it’s what I
0:30:43 know.
0:30:45 I worked for the Dodgers.
0:30:54 But also, because the value proposition of what we’re doing is so impactful, you try
0:30:59 to understand the relationship between people who have tried to put numbers on this.
0:31:03 People have estimated it’s in the millions of dollars, but the value of adding one mile
0:31:08 an hour of fastball velocity to a pitcher, people have valued that.
0:31:10 We have valued that in the millions of dollars.
0:31:17 Well, sure, you would value it in the millions of dollars, but what is the, I don’t know,
0:31:18 what’s a top pitcher make these days?
0:31:19 I don’t even know anymore.
0:31:25 I mean, you want to talk about Shohei Otani, half a billion.
0:31:26 Hundreds of millions, right?
0:31:29 Half a billion.
0:31:35 So right, so the marginal benefit has a very large value, like a million dollars against
0:31:36 a hundred million dollars.
0:31:40 One percent better if they’re making a hundred million dollars, that’s worth a million dollars,
0:31:41 presumably.
0:31:45 Yeah, we still have to get the pro sports teams to believe that.
0:31:51 Well, how many pro sports teams are paying you at this point about?
0:31:55 Close to 10 in Major League Baseball and a couple in the NBA.
0:31:57 The NBA is a lot newer.
0:31:58 And how big is the field?
0:32:03 Like what’s sort of the broader state of play in the field of, I don’t even know what to
0:32:04 say.
0:32:09 Biomechanics as a service seems like a niche construction of it, but what would you say?
0:32:10 Sports analytics?
0:32:13 I mean, I guess that’s not quite right.
0:32:16 How do you construct the broader field?
0:32:18 Sports analytics is definitely a part of it.
0:32:26 The data that we provide is novel or is different than traditional sports analytics.
0:32:31 So we don’t really have a lot of companies as competitors.
0:32:41 Honestly, our biggest competitors are teams wanting to try to do this type of thing internally.
0:32:45 Hire a bunch of data engineers, hire a bunch of software engineers, hire people with biomechanics
0:32:50 backgrounds and try to build out these processing pipelines themselves.
0:32:55 Does every team have somebody with a PhD in biomechanics working for them now?
0:32:56 In baseball?
0:32:57 Yeah.
0:32:58 Wow.
0:33:02 Basketball, not yet, but they’re starting to.
0:33:07 So I know people, I mean, I think in general, baseball fans like to complain.
0:33:12 But so one of the things they have complained about lately is the way analytics more broadly
0:33:14 made the game more boring, right?
0:33:20 Like the shift and changing pictures more frequently and whatever else people complain
0:33:21 about.
0:33:26 Are you, do you fit into that at all?
0:33:35 Oh, that’s a good question.
0:33:41 Some people complain about how hard pictures are throwing these days, because it creates
0:33:45 more strikeouts and people think strikeouts are boring.
0:33:47 So maybe, yeah.
0:33:49 So maybe you need to get better at helping hitters.
0:33:53 That way you can even it back up.
0:33:54 Right.
0:33:55 And that’s what we talked about.
0:33:59 Take aways for hitters are harder because they’re reacting to the picture.
0:34:05 So if you think about the field, what your company say in five years or whatever’s your
0:34:11 kind of medium term future that you think about, what is the company and the sort of
0:34:16 the world, the sports world that you’re interacting with look like at that time and say whatever,
0:34:18 five years, 10 years?
0:34:30 What I hope is that what we are trying to foster is an environment where coaches have
0:34:35 really incredible tools at their disposal to understand how an athlete moves.
0:34:40 So we talked about at the very beginning where sort of the still more or less the state of
0:34:45 the art is a coach just looks at an athlete, watches video of an athlete and tries to give
0:34:50 the athlete feedback regarding what they see on the video, what they see with their eyes.
0:34:59 We hope that the standard becomes you use an analytical tool to help you understand how
0:35:04 the athlete is moving and to really like level up your coaching because now you have this
0:35:08 objective information about how the athlete is moving.
0:35:13 I kind of make the analogy related to just radar guns.
0:35:18 Before radar guns were a thing, a coach would just like look at a picture and be like, “Oh,
0:35:26 that looks pretty fast, make an adjustment and I think that looks a little faster.”
0:35:29 But then radar guns came out and you could actually measure how fast the picture was
0:35:34 throwing and you could actually measure if the picture is getting throwing the ball harder
0:35:37 based on your feedback.
0:35:42 And now you’re doing that but in a way more complex way.
0:35:43 Exactly.
0:35:44 Yeah.
0:36:01 We’ll be back in a minute with the lightning round.
0:36:02 Let’s do the lightning round.
0:36:05 Let’s start with a few baseball questions.
0:36:09 Who’s the most underrated pitcher of all time?
0:36:12 Whoa.
0:36:16 Underrated pitcher of all time, oh my goodness.
0:36:22 I don’t know if I can give you one that’s all time because I don’t know if that’s fair,
0:36:23 I’m sure.
0:36:24 I’m not thinking of everybody.
0:36:25 Of your lifetime.
0:36:26 Of your lifetime.
0:36:29 I grew up a hardcore Red Sox fan.
0:36:32 I grew up in Rhode Island.
0:36:37 And the first one that comes to mind, mostly in the Red Sox atmosphere, but I wonder if
0:36:43 you could make a broader argument was Tim Wakefield, who was a knuckleball pitcher
0:36:44 for the Red Sox.
0:36:45 And what made…
0:36:46 Amazing physics.
0:36:49 Amazing physics and aerodynamics.
0:36:56 And the reason being is he just did so many things for the Red Sox, ate up so many innings
0:37:03 and was so effective closing, starting, whatever, but he never got incredible recognition because
0:37:07 it was a knuckleball that was going 55, 60 miles an hour.
0:37:12 What’s one thing you would change about baseball to make it more popular?
0:37:20 I do feel like actually the changes that are being made are good ones in reducing the amount
0:37:21 of downtime.
0:37:25 It’s what, a pitch clock is one?
0:37:26 Is that the way you’re thinking of?
0:37:27 Yeah.
0:37:28 Yeah, exactly.
0:37:29 Yeah.
0:37:30 A pitch clock.
0:37:33 The thing that’s challenging for me as a biomechanist is when you reduce the amount of time that
0:37:39 a pitcher has to throw, you theoretically could introduce more fatigue, which theoretically
0:37:41 also introduces more injury risk.
0:37:45 So this is something that we’ve been thinking about is like…
0:37:53 So for me, while I like that change that baseball is making to speed the game up, I think the
0:38:00 pitchers also need to train a little bit differently to be able to better withstand the shorter
0:38:03 rest time.
0:38:09 Are there aspects of your work and the changes you’ve seen over the course of your career
0:38:16 that illuminate sort of broader changes in computer vision and AI more generally?
0:38:17 Yeah.
0:38:22 In particular, the most important one has been the improvement in computer vision, because
0:38:29 computer vision at its core is artificial intelligence, neural networks, and as that
0:38:33 technology has gotten better and better, the latest and greatest, people always talk about
0:38:36 the transformer model really changed AI.
0:38:38 I mean, that changed computer vision too.
0:38:44 Lots of the modern, more modern computer vision models are based on transformers.
0:38:51 Which to be clear, the transformer model is what gives us chat GPT, the T in GPT is transformer.
0:38:54 So how has it affected the computer vision side?
0:38:58 Making the models more accurate and more efficient.
0:39:02 It used to be a couple of years ago when I tried to run a computer vision model on my
0:39:10 laptop, it could take an hour, just to analyze one pitch, one video, and now it takes a matter
0:39:13 of seconds.
0:39:20 And so what is the bigger implication of that beyond your work?
0:39:26 More and more data is available regarding how people move.
0:39:31 When I first started, it was really hard to have data in a baseball game.
0:39:36 Now every major league game, every minor league game, every NBA game, maybe every G-league
0:39:46 game, every WNBA game, every single basketball and baseball game, more or less, is now being
0:39:51 recorded with computer vision to get the three dimensional data about how people are moving.
0:39:57 Just lots and lots of data on how people move, and this is really impacting, I think, lots
0:40:04 of fields, in particular, I think, like self-driving cars, robots that are meant to interact with
0:40:07 the world, all rely on computer vision models.
0:40:12 I mean, I think one of the coolest things about how cars do this sort of thing is not
0:40:21 only do they have to understand where a person is now, but they’re really cool models that
0:40:28 they take where a person is now and the last 10 seconds of what that person did and try
0:40:31 to predict all the different things that the person might do.
0:40:36 They might run across the street, they might jump out of the way, they might jump forward,
0:40:38 they might run and chase a soccer ball.
0:40:44 I mean, my sense is all of the hard edge cases in self-driving cars, basically the reason
0:40:50 we don’t truly have self-driving cars yet, is because people are so hard to understand.
0:40:55 If the world was all self-driving cars, then it would be a solved problem, right?
0:40:59 The machine could understand what other machines are going to do, but people, human drivers,
0:41:08 human pedestrians are strange and very hard for machines to understand at the same time.
0:41:09 It’s like they need a coach.
0:41:13 It’s like the coaching problem.
0:41:18 We found your sound cloud.
0:41:21 So Jimmy Buffett fans are called parrot heads.
0:41:26 What are Jimmy Buffett fans called?
0:41:31 I’ve never thought about it.
0:41:38 You have a song called “Let’s Have Some Fruit”, parenthesis, the fruit song.
0:41:41 Is fruit a metaphor?
0:41:43 Leave it up to your imagination.
0:41:45 Fair.
0:41:50 Jimmy Buffett is the co-founder and CEO of Reboot Motion.
0:42:06 Today’s show was produced by Gabriel Hunter Chang.
0:42:11 It was edited by Lydia Jean Kott and engineered by Sarah Brugger.
0:42:14 You can email us at problem@pushkin.fm.
0:42:18 I’m Jacob Goldstein and we’ll be back next week with another episode of What’s Your Problem.
0:42:19 I’m Jacob Goldstein and we’ll be back next week with another episode of What’s Your Problem.
0:42:21 (upbeat music)
0:42:23 (upbeat music)
Jimmy Buffi is the CEO and co-founder of Reboot Motion, which uses biomechanics to help athletes in Major League Baseball and the NBA. Jimmy’s problem is this: How do you turn data about how professional athletes move into knowledge that helps them perform better?
This is the second episode of our series about people who are working at the frontiers of technology to help elite athletes perform better.
Music: Let’s Have Some Fruit (The Fruit Song) by J Buffi
See omnystudio.com/listener for privacy information.