🏇 Japan has built an AI that detects whip use in horse racing by sound. Race stewards currently count strikes by reviewing video footage by eye, a slow process that can miss things. Researchers at the University of Tsukuba turned instead to the sound a whip makes, producing a system that detects strikes with roughly 70% accuracy and may be fast enough to run live. It is an attempt to protect both the fairness of the race and the welfare of the horse.

Background: Why Automated Whip Detection Matters

The whip is a standard tool in horse racing, used by jockeys to encourage speed and maintain a horse's focus. However, excessive whip use can cause pain and distress to horses, making it a significant animal welfare concern that has prompted regulatory action worldwide.

The Japan Racing Association (JRA) tightened its rules in 2023, capping consecutive strikes at five without a two-stride pause, down from ten. Penalties escalate from a warning to fines and, for repeat offenders, riding suspensions. Other jurisdictions go further. Germany cut its limit to three strikes per race in 2023, and Britain reduced its caps the same year to six on the flat and seven over jumps.

Enforcement still relies on stewards reviewing race footage after each event. It is slow work, and it carries the ordinary risks of human error. What the sport has needed is an objective, efficient way to monitor whip use.

The Whip Sound Runs Past 22 Kilohertz

The research team, led by Associate Professor Keiichi Zempo from the University of Tsukuba's Faculty of Engineering, Information and Systems, took an innovative approach by focusing on the distinctive sound produced when a whip is cracked.

After collecting and analyzing sound data from various whip types used by trainers in stables, the team found that whip sounds contain frequencies above 22 kilohertz, past the range of human hearing (roughly 20 Hz to 20 kHz). Standard recording equipment cannot capture these accurately, so the researchers recorded at a 192 kHz sampling rate.

Building the AI Detection System

The team utilized deep learning technology called "Convolutional Recurrent Neural Networks" (CRNN), which can capture both the acoustic characteristics and temporal changes in sound patterns simultaneously.

Training data came from 24 races held in Japan, incorporating 620 instances of whip use. The best-performing model achieved approximately 70% accuracy in automatically detecting whip strikes. Crucially, the use of high-fidelity audio data containing high-frequency components improved detection accuracy, scientifically demonstrating for the first time that whip sounds contain these ultrasonic elements.

Real-World Testing and Real-Time Potential

The research team conducted field tests at racecourses in Japan's Kanto region and at France's ParisLongchamp Racecourse. Racing venues are acoustically brutal: crowd noise can reach 70 to 126 decibels, comparable to standing near a jet engine.

The team found that data collection was most effective in areas with fewer spectators, such as near track corners. In most configurations, processing ran faster than real-time audio playback, suggesting the system could detect whip use during live races rather than after them.

Implications for Animal Welfare and Racing Integrity

In practice, the technology would allow more accurate, objective monitoring of whip use, raising race integrity while easing the load on the horses. Zempo says he wants to develop it to the point where it helps reduce that load.

The research was funded by NEXION Corporation (Shinjuku, Tokyo). The results appeared on November 19, 2025 in "Engineering Applications of Artificial Intelligence," a journal of the International Federation of Automatic Control (IFAC), and the University of Tsukuba announced them on December 12 of that year.

The Global Context: Growing Focus on Animal Welfare

The debate over whip use continues worldwide. Harness Racing Australia decided in December 2016 to ban whips outright and did so from September 2017, the first time a racing authority gave up the whip on its own initiative. Norway has prohibited whip use in ordinary racing since 1982 under animal welfare legislation, though riders may still carry one.

Jockeys remain wary. Yutaka Take and Christophe Lemaire have both objected to tighter limits, arguing the whip is a signal for safety and steering rather than a device for inflicting pain. What this technology adds to that argument is objective data.

Looking Ahead

The research team plans to keep collecting data and refining the system so it works reliably in noisy conditions, with practical deployment as the goal. If it gets there, the argument over how many times a horse was struck moves from impression to measurement.

In Japan, AI technology is being developed to promote fairness in horse racing while improving animal welfare. What discussions are happening in your country about how animals are treated in horse racing or other sports? We'd love to hear your perspective.

References