Using virtual reality to boost skills in Olympic boxers

The 2024 Olympic Games are just around the corner and it’s not just the city of Par­is that’s get­ting ready… sci­ence is get­ting involved too! At Inria Rennes, we are con­duct­ing research with the Olympic box­ing team and its coach: as part of the REVEA pro­ject¹, we are look­ing at how vir­tu­al real­ity can help the team work on cer­tain “sub-skills” that need to be improved, such as speed, motor coordin­a­tion and strength train­ing, to name a few. Con­trary to what one might expect, our vir­tu­al real­ity sys­tem does not aim to rep­lic­ate a real match or com­pet­i­tion, but rather to identi­fy a spe­cif­ic sub-skill and improve it in a tar­geted com­bat situ­ation. In the case of box­ing, the sub-skill we are look­ing to train is the abil­ity to anti­cip­ate oppon­ents’ attacks.

To ana­lyse the oppon­ent’s beha­viour without vir­tu­al real­ity, two box­ers must com­bat face to face, with one of them being asked to repro­duce typ­ic­al attacks. It is very dif­fi­cult, how­ever, to con­trol and faith­fully repro­duce cer­tain move­ments from one tri­al to the next. This approach inev­it­ably lim­its train­ing time since such an exer­cise is phys­ic­ally demand­ing when real punches are being exchanged. In vir­tu­al real­ity, this type of train­ing can be done with our avatar and a vir­tu­al oppon­ent without the risk of any­one being injured. It also offers a range of stim­uli that go bey­ond what is pos­sible in real life: we can exper­i­ment with vari­ous visu­al and aud­it­ory per­cep­tions or prac­tise with a highly-skilled oppon­ent to push the lim­its of tra­di­tion­al training.

To this end, our team has developed a vir­tu­al real­ity sim­u­lat­or that uses three-dimen­sion­al meas­ure­ments of the move­ments of pro­fes­sion­al box­ers obtained with motion cap­ture devices – via the Immer­move plat­form installed in the M2S labor­at­ory) The data obtained from this sim­u­lat­or, which includes, for example, anti­cip­at­ory defens­ive move­ments, such as slight hip move­ments, is then trans­posed onto vir­tu­al humans with dif­fer­ent mor­pho­lo­gies. As such, the box­er can visu­al­ise his oppon­ent, rep­res­en­ted by an avatar and dis­played in a vir­tu­al real­ity head­set. In our exper­i­ments, we observe the box­er­’s reac­tions to dif­fer­ent types of blows, and wheth­er he is sens­it­ive to the slight anti­cip­at­ory move­ments per­ceived by the oppon­ent. These obser­va­tions can be quan­ti­fied, for example, by suc­cess rates or reac­tion times, to cre­ate a “map” of the box­er­’s anti­cip­at­ory performance.

As well as apply­ing it to box­ing, we used our approach to train goal­keep­ers at the Stade Ren­nais Foot­ball Club. Our vir­tu­al real­ity sys­tem trained them not to focus on the play­er with the ball, but to bet­ter man­age peri­pher­al vis­ion inform­a­tion. Indeed, if the foot­baller about to attempt a goal makes a pass to someone on the oth­er side of the goal at the very last second, the goal­keep­er may not have time to react.

Our research is very applied: our object­ive is to help the Olympic teams get ready for Par­is 2024. Indeed, think­ing about how to include these tech­no­lo­gies in the daily life of ath­letes is one of the main goals. We asked ourselves the fol­low­ing ques­tions: how much train­ing should be done with this type of tech­no­logy? At what time of the day and at what stage in the train­ing pro­gramme? How long should a typ­ic­al ses­sion last? Who should we train with – Olympic hope­fuls or all ath­letes, for example? And how often? Our approach is part of a wider train­ing pro­gramme set by the coach, which includes phys­ic­al, men­tal and tech­nic­al preparation… 

The train­ing pro­gramme includes phys­ic­al, men­tal, and tech­nic­al pre­par­a­tion… with added vir­tu­al real­ity sessions!

Our sys­tem is being used at the Olympic team’s train­ing centre and by the sci­ent­ists at INSEP work­ing with the French Box­ing Fed­er­a­tion. We are con­tinu­ing to devel­op it and make it more effect­ive thanks to the feed­back we receive from ath­letes and coaches. For example, in our first ver­sion, the oppon­ent made too many repet­it­ive move­ments, which made the com­bat unreal­ist­ic. We are also improv­ing the equip­ment used for vir­tu­al real­ity training. 

We have a soft­ware plat­form that is based on a com­mer­cial tool called Unity, routinely used for mak­ing video games. It loads a three-dimen­sion­al envir­on­ment – in our case, a box­ing ring – as well as vir­tu­al humans and their move­ments. We write a scen­ario in which the move­ments of the vir­tu­al humans are stored in “move­ment lib­rar­ies”, which we play back. There are then algorithms linked to the motion con­trol of the avatars. All these ele­ments provide us with the met­rics that allow us to meas­ure the reac­tion times and the anti­cip­at­ory beha­viour of the boxers.

We have also star­ted to work on arti­fi­cial intel­li­gence, neur­al net­works and deep rein­force­ment learn­ing to improve the beha­viour of vir­tu­al oppon­ents. This part of the tech­no­logy is not yet mature, but ini­tial res­ults show that it is pos­sible to repro­duce the com­plex inter­ac­tions between two box­ers with very few obser­va­tions of real com­bats. Tak­ing this approach to its lim­its, we could con­sider sim­u­lat­ing the move­ments and strategies of a par­tic­u­lar box­er based on some video foot­age of his pre­vi­ous matches.

Interview by Isabelle Dumé