The emergence of new technologies has seen sports organisations implement AI to seek a competitive edge on and off the playing field, deepen their understanding and accelerate the race for information to help steer their future.
Currently, organisations are striving to adopt advanced practices to improve officiating, identify talent and ultimately save time and cut overheads. It isn’t an easy win, as our various use cases illustrate, the benchmark for quality is high, and we humans are more tolerant of human error than we are of technical malfunctions. As manifested weekly by many tennis players and Premier League managers.
Over the past five years, we have witnessed the introduction of AI features at the highest levels in sports, such as tennis, with the implementation of Hawkeye technology. In certain tennis tournaments, AI is now being used for line calls during matches, replacing traditional line judges. Despite its benefits, this approach has sparked controversy and debates as AI’s interpretations are not 100% accurate, which poses a large problem in high-performance sports as incorrect conclusions can be the difference between winning and losing.
Crossing the Line
At Wimbledon in 2022, GB’s Joe Salisbury and partner Rajeev Ram demanded that Hawkeye be turned off during their match and even refused to play on momentarily due to a disputed line call. Furthermore, world number 5 Jessica Pegula once recalled a match where the automated line call was incorrect and the umpire told her “Actually you are correct, Hawk-Eye was wrong. The ball was out.” Ball tracking technology is also yet to be implemented during clay court tournaments, with the systems accuracy threshold not high enough on the surface.
A problem arises here whereby the players may not trust the systems being used, and if a sport with such large funding as tennis cannot be accurate 100% of the time, newer, emerging sports are facing an even steeper challenge.
Padel is the fastest-growing sport in the world, and like other sports, athletes and coaches in the space are keen to develop AI to produce data quickly and accurately. However, here some of the challenges and limitations of generative AI become apparent, are with ball tracking accuracy and shot identification, due to the complexity of the sport.
Beyond the Scoreboard
Numerous studies have explored the potential of machine learning algorithms and AI technology in generating padel data. These studies have utilised various technologies, ranging from wearables to video tracking, but each method comes with its limitations. Especially when the technology can’t leverage a human-operated scoreboard or crowd noise to error-check their data.
- Javadiha, Andujar and Lacasa (2022) - Object tracking from video of padel matches – Discovered the system they use cannot account for shot effects such as spin. They concluded that tracking software must become more robust and accurate, and specialised tools will be required to fully analyse padel tracking data.
- Svensson and Hult (2022) - Person identification in padel - Concluded that unknown and unseen influences affected their model which was specifically designed for padel, and scenarios arose where the net interfered with player ID, as well as clothing and re-entering court.
- Dominguez, Alvarez, and Cordoba et al (2023) - Deep-learning algorithms for stroke classification using wearables - They were able to classify 13 different strokes, but padel is a much more layered game than tennis, with the nuances of many shots varying depending on their spin and speed.
The best algorithm tested yielded an accuracy of 93.35% for shot identification, which initially seems like a good result, however, the margin for error is far too high to be trusted. If we were to take a standard rally length of 10 shots, the likelihood of each shot alone in the rally being accurate would be 50%.
Finally, as of today, AI in padel is simply not advanced enough yet to be used at the top end of the game. The variation in techniques, playing styles and the abundance of shot types means that expert analysts manually creating data is more reliable than any AI technology. At Skylab, we use performance analysis software to create data on a variety of aspects of padel, all produced by a human expert with high levels of knowledge of the sport.
The use of AI may be suitable for recreational players who want to add some numbers to their game for healthy competition, however, in elite padel with professional and developing athletes, it cannot be relied upon in its current form to produce valid and meaningful information.
These limitations aren’t isolated to sport, it was recently revealed that Amazon’s ‘just walk out’ technology, which allowed shoppers to pick up items and leave without going through a checkout, was only actually implemented in half of their stores, and that the other half were relying on workers in India tracking individuals and their purchases remotely, feigning the AI technology.
Looking forward, Skylab is determined to continue to research and develop AI technologies into day-to-day activities combined with best practices in Performance Analysis. This means the accuracy, robustness and reliability of the data are non-negotiable and until then, we will prime ourselves to be ready for the day technology will be fit to implement and accessible to all rather than just the lite, which will take our field into a whole new dimension.
References
A query language for exploratory analysis of video-based tracking data in padel matches - Javadiha, Andujar and Lacasa (2022)
Person re-identification in the sport of padel - Svensson and Hult (2022)
A comparative study of machine learning and deep learning algorithms for padel tennis shot classification - Dominguez, Alvarez and Cordoba et al (2023)