A New Stroke

Why Australian Crawl is on its last laps

June 5, 2024

Up-and-coming swimmer Bada Choi, along with his unorthodox support team, is poised to disrupt the entire sport of freestyle swimming. We investigate Bada, the minds behind his success, and the novel machine-learning code that has enabled it.

There is chlorine in my nostrils and puddles underfoot in the Aquatic & Fitness Centre, at the heart of the Australian Institute of Sport in Canberra. Bada Choi—tall, tan and lithely muscled in a black speedo, yellow cap and electric-blue goggles—stands at the end of the pool swinging his arms. He catches my eye and grins. There’s nothing in his demeanour, as he steps up to the diving block, to suggest anything unusual about this training session. But the normalcy of the scene belies the novelty of the cross-discipline collaboration that is underway here, and the waves it has sent crashing through Australian sport.

Behind Bada, whistle and timer in hand, is Dr Blake Sidana-Carrington, sport scientist. Seated on the white benches to the side of the pool is the computer scientist Dr Julia Berta. On Julia’s lap, commanding most of her attention, is the item most crucial to the swimmer’s success: an open laptop, containing, somewhere on its drive, a branch of the Roshi Algorithm.

The pool is already swarming with activity. Swimmers are training in the other lanes, soaring rhythmically along the pool in various states of energy. The Paris Olympics are scant weeks away, and there isn’t a minute to waste. I note breaststroke, butterfly, backstroke, and of course Australian crawl—the style more commonly known as ‘freestyle’, after the event in which it is universally deployed. I note a look or two directed Bada’s way from those preparing for their dives. It’s difficult to read facial expressions through goggles, so perhaps I am projecting the hint of resentment that I seem to detect.

“Take your mark,” barks Sidana-Carrington, with the harsh staccato of swimming coaches everywhere, lifting the whistle to their mouth. Bada hunches over and grips the front of the diving block, his body shedding relaxation for spring-loaded readiness. Blake blows a sharp note on the whistle. Bada launches himself from the pad, arcs gracefully through the air, and pierces the water with a quiet splash. He kicks twice with both legs together, angling toward the surface with the same sublime elegance as his dive—a quality utterly undermined when he surfaces to commence his stroke.

The style Bada begins to swim calls to mind the mad scramble of a huntsman spider as it dashes away from a pegged thong; it would be hard to believe that his thrashing provides much thrust at all, were it not for the other evidence of my eyes. Much like the huntsman, perhaps the most astonishing aspect of Bada’s bizarre movement is the speed with which it propels him through the water. Dr Sidana-Carrington follows Bada along the poolside, their long strides not quite keeping pace with the swimmer. I have to jog to match him, notebook hanging forgotten by my side. I’m astounded. I’ve seen, of course, the tapes of Bada’s now-infamous Australian Swimming Championship races, but watching him work in person is something quite different. Despite having surely witnessed this sight several times before, a couple of the other swimmers—Olympic qualifiers themselves—seem to agree, pausing to turn their heads as he flails by. After an interval of shocking brevity, Bada’s hand taps the pool wall. Blake hits stop on the timer. A grin spreads across their face, and they show me the numbers on the screen.

Until broken in Tokyo by Caeleb Dressel, the Olympic record for men’s 50-metre freestyle was 21.30 seconds, set by Brazilian swimmer César Cielo during the controversial Beijing Olympics. Bada has completed his warm-up lap in 21.29—and without the aid of the notorious super-suit.

It might do here to pause and recall precisely what makes Bada so unique as a swimmer, and the importance of his non-swimming collaborators. All three members of the unlikely team are relatively young, and all are presently affiliated with Macquarie University, Sydney. It was here that this peculiar collaboration was born. Bada is 20, partway through a deferred undergraduate course in International Studies. Doctor Sidana-Carrington, at 28, graduated just over two years ago from his PhD at Curtin University and is almost two years into his first post-doctoral fellowship. Dr Berta, even younger at 26, is most of the way through hers, having obtained her PhD at the very institution she is currently affiliated with. And now, this trio is poised to disrupt the entire sport of freestyle swimming.

Australian crawl, also known as ‘American crawl’, ‘front crawl’ or simply ‘freestyle’, has long been considered the fastest and most efficient swimming stroke. Watching the Olympics, you might be fooled into thinking that ‘freestyle’ is the true name of the technique—but this is not so, as Dr Sidana-Carrington explains.

“You don’t actually have to swim crawl in the freestyle race—you could swim butterfly or breaststroke if you wanted. That’s why the event is called ‘freestyle’. It’s just that crawl is the one everyone uses. That’s all about to change.”

One might believe that front crawl has always been with us, such is its ubiquity. Although relatives of the style have probably been in use since ancient times—there are claims of portrayals in ancient Egyptian art—a version of front crawl first splashed into the western world in the mid-1800s, when a pair of Ojibwe swimming champions—indigenous Americans from the land straddling the modern border between Canada and the USA—were invited to an exhibition in London. They swam the stroke they knew best, flying up the 130-foot pool inside of 30 seconds. English observers were left clutching their monocles, sputtering objections about the ‘un-European’-ness of the swimmers’ ‘grotesque antics’. Hence, chauvinism managed to quash the stroke’s adoption, despite its obvious speed advantage, and breaststroke remained standard for some further decades. Sidestroke enjoyed a brief moment in the spotlight after it was copied from indigenous Australians in the Sydney region, while the hybrid ‘trudgen’ stroke was developed by imitating Black African swimmers in the 1870s, surging in popularity due to its improved speed. It was not until Solomon Islander Alick F. Wickham rose to fame in Australian swimming in the 1920s, using a variation of the crawling technique popular in his homeland, that the Australian crawl found its modern name. White Australians set off to quietly copy Wickham’s style, and within decades it had become standard in western competition. Since then, average lap times have fallen incrementally, as swimmers gradually improved their techniques and their training methods. It seemed that the time for true innovation in swimming technique had passed. Since then, as the fastest stroke, there has never been any reason not to use Australian crawl when permitted. Or, at least this was the case—right up until Bada Choi dove headfirst into the Australian Olympic team.

The stage is now set for a revolution in freestyle times, ushered in by Bada’s new technique—tentatively, and descriptively, dubbed ‘flailstroke’. Just as western swimmers were once shocked by the jump in speed offered by the front crawl, the international swimming community has been taken aback by flailstroke, and the otherworldly gains it offers. However, unlike crawl and its predecessors, flailstroke is not an appropriated version of a style practised, in seclusion from the west, for centuries—it appears not to have existed at all until the Roshi Algorithm happened upon it.

“It’s not a new type of Crawl, or butterfly or anything else. It’s a whole new stroke, a whole new family,” Blake explains excitedly from the bench. Julia Berta’s eyes seem not to have risen throughout Bada’s lurching lap; her hands haven’t stopped flying over her keyboard, the onscreen wall of bright monospaced lettering morphing swiftly against her screen’s dark background. Although the Algorithm itself is firmly within her wheelhouse, Blake takes the lead in explaining its success.

“It turns out Australian crawl is located in what mathematicians call a ‘local maximum’. That’s an efficiency peak that’s higher than the immediate surroundings, but not necessarily a maximum overall. All those decades of refinement were just pushing it to the best it could be within that bubble. Major variations from the normal stroke only gave declines in efficiency—or that’s what everyone thought. We got the Algorithm to jump far away from front crawl, to look for new maxima in the simulations. And it found them! We just had to get out of that local bubble.”

Sidana-Carrington and Berta share credit for their discovery with Roshi—their joint brainchild, a genetic machine-learning algorithm tuned to explore the movement of the human body. The two scientists arrived at the name ‘Roshi’ together; both are confessed anime geeks, naming their creation after a character from the popular Dragon Ball franchise. The algorithm searches a wide range of swimming styles using an initial pseudo-random scattering in a multidimensional swimstroke-space, each stroke varying greatly in the movements of the arms, legs and torso. Unless the initial seed is lucky indeed, none of this first generation is any good at all; Sidana-Carrington showed me a handful of animations, almost comical in their limp ineffectuality. In one, the computer-generated swimmer furls and unfurls like a caterpillar crawling along a leaf, at a rate measured in metres per hour. But from the first generation, a set of variants with slight, randomised differences is generated. The algorithm churns through these permutations, examining each in simulation. This simulation, involving a sophisticated model of the hydrodynamics and biomechanics of the human body—including the need for the swimmer to breathe—was conceptualised by Dr Sidana-Carrington and coded by Dr Berta, with feedback flowing in both directions. The performance of each stroke variant is evaluated using a ‘fitness function’, and a batch of new permutations is developed from the best-performing while the rest are discarded. This is the ‘genetic’ component of the algorithm, a computational process that uses randomised, generational variation in traits and repeated tests of performance to ‘evolve’ its subject toward better efficacy—just like natural selection in the biological world. Although it might seem inelegant at first glance, Roshi’s approach is not random, or even truly brute-force.

“That’s where the machine-learning component comes in,” Julia Berta is happy to explain, closing her laptop for the first time since I arrived. Usually the most soft-spoken of the trio, Berta seems to enjoy little more than writing and executing code—the quiet rush of power she experiences when a program translates ideas into reality in a cascade of elegant, interlocking digital machinery. Occasionally, however, her apparent reticence bursts forth into eloquent streams of exposition.

“The algorithm learns how to pick up on local trends in fitness and trace them upward. That’s really the major innovation here, although the mass stroke simulation is certainly novel. The learning saves a great deal of computation, better-directs the focus. Otherwise, our walkers might be wandering for years. The best part, though, is that this MLA [machine-learning algorithm] doesn’t require a training dataset. That’s of vital importance for this application, which obviously doesn’t have much prior data lying around.”

The Algorithm extends each line of descent until there is nowhere left for it to go; once it has identified a local maximum, the evolution of that variant is halted. Most of these trails led right back to front crawl; there were several novel failures, too, local speed peaks that were nonetheless slower than crawl. But, after thousands of iterations, a handful of the walkers converged on an unfamiliar style, one that seemed to be, at least in theory, superior. Once the style was identified as proximal to a maximum, further simulations were applied at a higher fidelity to refine it, always in comparison to front crawl.

“After the simulation phase, we knew we needed a real swimmer to learn the stroke,” Sidana-Carrington continues.

“We were confident in the simulation. We weren’t sure, though, how well it would translate to a real human. We talked about building a robot to demonstrate the stroke physically, compare it to a crawl control, but that’s light-years beyond our budget. And they don’t let robots compete in Olympic events just yet. So the next step was to convince someone to learn the damn thing.”

That would mean months of commitment to training in this stroke and this stroke alone, not a burden that a professional swimmer can take on lightly. To the swimmer, it would mean falling behind in regular training to learn something completely foreign, without any real guarantee of a payoff. It would require someone willing to take a leap of faith. Enter Bada Choi.

Bada, born and raised in Sydney, has always been a strong swimmer, winning grade champion at numerous school swimming carnivals and working his way up to state-level competitions on three occasions. However, he had never qualified for a national event, let alone an international one. That changed in April, with the Australian Swimming Championships—which, during leap years, double as the qualifying event for the Australian Olympic swimming team. When Bada, a relative unknown on the scene of Australian swimming, beat the nearest qualifying time by an entire second, it tipped the global swimming community into uproar. But Bada owes his recent chain of success entirely to the new stroke, a fact he makes no pretence about.

“Oh, yeah, I wouldn’t ever qualify for the Olympics if it weren’t for flail, not even close. I always wanted to, though. I love swimming. I feel great whenever I do it, makes everything seem better somehow, even when I’m deep in the pits. I always worked hard at it, ever since primary—I had, you know, a picture of Ian Thorpe pinned to my wall—but I’m not really in the same league as the guys on the team. I came to terms with that a long time ago. I still love it, but it since high school it was more of a hobby. But then Blake comes up to me at the uni pool. They’d been watching me do laps—I was really pushing myself that day, working out some mid-sem exam stress. I thought they were interested in me for, you know, my body or whatever.”

But Bada had caught Sidana-Carrington’s attention for a somewhat less flattering reason.

“They thought I looked like a good enough swimmer to execute on their idea, but mediocre enough to be interested in taking a risk on it.” He grins at the memory.

“Didn’t tell me that right away. Got it out of them with soju after the Championships. I don’t care; they were right. I just feel honoured to be on the brink of something new. I’m making history. It could have been anyone else, I was just in the right place at the right time.”

Dr Sidana-Carrington has acted as Bada’s personal trainer ever since that first fateful encounter. They could—perhaps should, given the intense time demands of postdoctoral academia—have handed these duties off to a regular (and, in some ways, more qualified) swimming coach some time ago. But they’re determined to see this through, going so far as to brush off any suggestion that they have invested more in this project than necessary.

“Bada was the one taking on all the risk here. He really went out on a limb, committing himself like this. He had to train to move in a whole new way. It’s a tribute to the guy’s persistence. There was a while there where we were beginning to think this whole thing was a waste of time. But Julia kept pushing us, and Bada kept at it, and it all paid off.”

The risk was high, but the potential reward was much, much higher. After a year and change of daily training in the new stroke, Bada is at the top of the field.

Until April, the specifics of flailstroke were a closely guarded secret. Bada’s training sessions had been strictly closed-set, taking place after hours by special arrangement at the university pool. The first time anyone outside his circle had seen the stroke in action was at the Championship. Since then, trainers, swimmers and pundits alike have pored over his races frame by frame, attempting to deconstruct exactly what it was he was doing to go so damn fast.

“It’s this really weird, y’know, scramble,” Bada says, in an attempt to explain the experience of swimming the stroke properly—an experience which, thus far, he alone in the world understands.

“I wouldn’t say it’s the most energy-efficient stroke, not compared to freestyle, sorry, crawl; at the end of each race, I can barely climb out of the pool. But it’s sure a shitload faster, once you know how to use it.”

It took Bada weeks of training before he was able to swim a single lap of flailstroke, and months before he was able to break his front crawl personal best; the Olympics commence, at the time of publication, in mere weeks. Now, far too late for the competition to adopt the stroke, Bada is the firm favourite for every freestyle event in Paris.

Once Bada had broken the crawl barrier, his flail times shrank steadily, asymptotically approaching his current sub-20-second best. Around here they have hovered—and Bada feels will most likely stay. By 2028, Bada, by his own admission an above-average but below Olympic-standard swimmer, will be quite obsolete.

“Oh, yeah, by L.A. I’ll be left in the dust. It’s awesome to be a part of it, though! I’m going to the [expletive] Olympics! I’m probably gonna clean up in the freestyle events this year, and I’ll be in the relays, but that’s it. That’s all I qualified for! My other strokes aren’t close to Olympic level. And in four years this one won’t be either! I’m just riding the wave while it lasts.”

The element of surprise in the team’s strategy has been at the heart of the current controversy; the new stroke has been announced far too late for the rest of the Australian team to adopt it for Paris, much less the rest of the world. This strategy has not only allowed Bada a term in the spotlight; it has propelled flailstroke into the public eye in nearly the highest-profile manner conceivable—and, I am certain by no accident, advertised the success of the Roshi Algorithm.

Although Bada seems quite content with his four minutes of fame, there has been talk of veiled resentment among the rest of the Olympic team. When asked about this, Bada shrugs.

“Y’know, I probably shouldn’t talk about it too much, but yeah, there’s a bit of that. I’ve felt like a bit of an outsider, for sure. Dirty looks in the change room, that kind of thing. Nothing like real harassment, just kind of this cold feeling. But, you know, I can kind of see their perspective. They work their asses off for years, eyes set on the Olympics, completely, you know, fixated on this one event, especially after how weird Tokyo was; and then this guy comes along with what seems like a cheat code, just waltzes in. I don’t think that’s totally fair, I’ve worked hard all year too, but that’s how it looks to them. There’s one guy who didn’t qualify for the 50-metre because of me. I bumped another guy off the freestyle relay, one more off the medley. I’d be a little pissed too.”

qualify for the 50-metre because of me. I bumped another guy off the freestyle relay, one more off the medley. I’d be a little pissed too.”

“He’s too modest. These other swimmers don’t realise that they’re in the past. Bada is the future. He was the one who broke the mould.”

Despite her apparent inattention, the pride she has for Bada shines now nearly as brightly as that for her Algorithm. The trio, regardless of their disparate roles and academic origins, present an image of a tight-knit team, a united front. Between my questions, they keep up a steady stream of banter, with Julia trading barbs and in-jokes without glancing away from her laptop. There isn’t much reason for Julia to be here, at this late stage; her part was in the computational side of things, long since complete, but she has made an effort to visit Bada regularly. It is fortunate for Choi that he has this kind of support; this experience must be isolating for him.

The Dolphins—Australia’s national swimming team—have always been strong internationally, punching well above the small country’s weight and rivalling even the US team. But flailstroke is indeed set to revolutionise the team’s performance.

“Every swimmer who’s seen Bada has talked about switching up their stroke. There won’t be any room for crawl for much longer.”

Once the Paris Olympics are over, Sidana-Carrington is convinced other teams will be well on their way to reverse-engineering the stroke. By L.A. 2028, if everything proceeds as envisioned, the entire field of every freestyle race will be swimming flailstroke. However, given the difficulty involved in mastering it, Blake believes the Australian team will retain a competitive edge through the next Olympics. Some other team-members have already tried the stroke, Sidana-Carrington tells me, but all have given it up, at least until Paris.

“After that, I’ll be happy to train anyone in Australia to swim flail. But they’ll have to hire me first,” Dr Sidana-Carrington says, with a wink and a grin.

“As people train in the stroke and refine their techniques, I bet we’ll see the records creep down for years yet.”

The expected reductions might not seem like a huge leap to an outsider, but the new stroke’s impact will likely be even greater—and more permanent—than the use of tight suits at Beijing 2008, the root of a huge controversy in the sport. 25 world swimming records were set in Beijing, and 65 Olympic records, a massive step up from Athens’ respective 8 and 25, and the 9 and 25 earned in London four years later. This spike in swimming speed was blamed squarely on the LZR Racer, the buoyant, compressive, water-repellent suit that became ubiquitous shortly before Beijing—at least for those teams who could afford it. Although the initial number of new records will not be of quite the same magnitude—the tight suits were used in every event, while flailstroke is limited to freestyle—the quick succession in which they were scored may yet be emulated.

The suit, and others like it, were later banned by FINA (Fédération Internationale de Natation), the Olympic-recognised governing body for swimming. Flailstroke will be different; entirely, unambiguously legal and unlikely to be legislated against by any governing body. The ultimate performance of the stroke is, after all, a function of the athlete’s physical prowess. This was the premise of FINA’s 2009 ruling against the LZR suit, seeking to value physical performance over technological innovation. Although flailstroke might seem a grey area—it is, after all, the direct result of new technology—it would be odd indeed to ban swimmers from moving in a particular way, in an event that is, after all, named freestyle. Now that the cat is out of the bag, commentators seem to agree, it would be foolish to try stuffing it back in.

Flailstroke isn’t just a breakthrough for swimmers or merely for sport and computer science—it represents a victory for cross-discipline research.

“You don’t see very much inter-department collaboration in Australian academia these days, and even less at our university. There’s a good deal of indirect incentivisation against it; papers published in journals outside of your field don’t count toward your ERA score or your H-index. It’s a massive shame. If this whole business has demonstrated anything, it’s that the different sciences have a lot to learn from each other.”

It is Julia who expresses this to me, with Sidana-Carrington nodding alongside her. Both of the scientists have had to complete this work outside of their normal hours. Under the current academic system, despite the obvious success of the work, it would not count toward their employment at Australian academic institutions. Both Berta and Sidana-Carrington feel that this system engenders a certain staleness, a self-circulating anti-innovation bubble. They—unsurprisingly, given the benefits they’ve recently reaped—believe that if scientists were incentivized to engage more with other disciplines, the cross-fertilisation of ideas to follow would be a great boon to science and technology. I feel myself agreeing with them.

Dr Sidana-Carrington is cautiously optimistic about the algorithm’s application to other sports.

“With things like running, it’s believed that what’s currently done is peak efficiency. No room for improvement except in fitness and strength. But that’s what we thought about swimming, so who knows? We’ll have to get to work on new sims to test that. Actually, Roshi is still running the swimming program, on Julia’s office server. It might find an even faster stroke one of these days. Not super likely, but not impossible.”

Dr Berta is less measured in her enthusiasm.

“I don’t see any reason we can’t overhaul every sport. I think this is the beginning of a revolution in sporting technology. And it might just be a computational one, too,” she finishes, with a smile that I can’t help but read as coy.

Berta has been reluctant to elaborate on precisely what kind of machine learning powers her creation, the source code to which remains under careful guard. No wonder as to why—if they can be applied commercially, the algorithms she has developed have the most potential for material profit of any component of this endeavour. If, as Berta claims, the approach doesn’t require training, then it is unlikely to employ a neural network, the type of deep-learning popularly used for pattern recognition in data science and big business; at least, it is certainly not any class of neural network in the public literature. Perhaps it finds an innovative combination of a neural network with one or more other techniques, such as principal component analysis or random forest decision-making. Possibly, it is something entirely new, a computational technique as novel as the swimming stroke it has produced. The lack of a training requirement may prove revolutionary on its own merit. Short of flailstroke being a fluke, Roshi has thoroughly proven the effectiveness of the mystery technique.

Whether or not Roshi goes on to reinvigorate other competitive sports, or whether its success propels a new wave of scientific cross-innovation, it has made an indelible mark here and now on Australian swimming, and propelled the three people at the heart of its story into worldwide notoriety. I look keenly forward to following their journey as well as that of their creation, as its consequences—foreseen and unforeseen—take the world by storm.

Written by Lachlan Marnoch, 2018 - 2021