Unpredictability was once a player’s greatest weapon. Not anymore.
Imagine if you knew where your opponent was going to hit the next ball.
Imagine if you could anticipate their every shot, not by watching their body, but because an algorithm had fed you the data.
This is the future of tennis. In fact, this technology is already here.
The skill in sports such as Tennis is dependent on the speed, accuracy and agility with which players respond to an unpredictable stimulus like a sneaky drop shot or a swift volley.
But a new algorithm could change the predictability equation by helping players foresee where their opponent might hit a ball.
A group of researchers from the University of Technology in Queensland analysed thousands of shots Novak Djokovic, Rafael Nadal and Roger Federer made during the 2012 Australian Open tennis tournament.
Tennis Australia provided the data from the Hawk-Eye review system which is used during matches to review shots.
They used this information to predict how players might respond to different conditions during a match.
“We wanted to predict trajectories and model decision making,” Dr Simon Denman, Senior Research Fellow in the Speech, Audio, Image and Video Technology Laboratory told 10 daily.
This is achieved by feeding the Hawk-Eye data into the algorithm to generate a prediction based on the position of the ball and the player, game score, ball speed and the path of the ball.
Once the data is given to the machine, it uses learned processes to mimic what a player’s brain might be doing during a match. This is achieved with the use of two different memory types — the episodic memory and the semantic memory.
“The episodic memory says to the algorithm, ‘have you seen anything like this?’ and can basically go through all the previous shots it has seen,” Denman said.
It then aggregates all the information that appears similar to the shot it has been fed to predict what could happen with the next shot.
The semantic memory holds more generalised information.
“It is capturing more higher level, semantic learnings about tennis like more general information, rules of the game, these sorts of things… more high-level learning than we would get from watching tennis for many days, hours and weeks.”
The memories combine their information to make a prediction on where the next shot will go and the algorithm then produces a pictograph so players and coaches might use it to learn from.
“It comes down to being able to better plan for your upcoming matches,” Denman said of the algorithm’s practical applications.
“If you know who you are going to play next, you can start to analyse how they tend to respond in different match situations and you can start to come up with your own strategies to counteract what they are going to do.”
Denman and his team say the algorithm could be used for match preparation and coaching purposes within the Tennis world because of the speed of the game.
“Tennis is a pretty fast moving sport, so by the time you get that prediction it has already happened.”
Denman believes this kind of technology will be available to top-level players within the next decade. The hardest part of using the new tech is obtaining the data in the first place. It is up to the Tennis tournament or the body who owns the data to release it to the public.
“We got it [the data] provided by Tennis Australia. It is not something that is available publicly,” Denman said.
“Perhaps that is the biggest challenge in getting this technology into the mainstream.”
For competition to be fair, all players would have to have their data released. For this to work, sports authorities would need to come to an agreement with players about how data would be released.
Denman also said the algorithm’s applications could easily spread beyond tennis.
“The model would need to grow to take on other sports like Cricket to try and predict where a batsman is going to hit a ball or where a bowler is going to bowl a next delivery.”
“These things are all possible if we have data.”
Featured Image: Getty Images.
Contact Siobhan at networkten.com.au