Miwa Harimoto, at just 17 years old, has already become the world’s top non-Chinese female table tennis player. In the next Olympic cycle leading to Los Angeles, I expect her to rise even higher and become a serious challenger to China’s dominance. Some pessimists predict she will follow the path of her predecessors — Mima Ito and Miu Hirano — who peaked early and declined as they grew older. I disagree.

Mima Ito’s style is unique but once decoded, she loses her advantage against Chinese players. Miu Hirano’s physical condition cannot sustain her ultra-aggressive approach, so her decline is understandable. Miwa Harimoto is different. As the second-tallest player on the Japanese team, she has the physical foundation to continue developing. Her loop-drive fast-attack style, played with double-sided inverted rubber, embodies the philosophy of slow build, strong release. Her success will come from deep preparation, and she has not yet reached her peak. She also benefits from exceptional training partners — her parents, both professionals, and her older brother Tomokazu Harimoto, currently ranked №4 in the world.

While mimicking Miwa’s strokes in my own training, I came across a video where she was invited to test a new robot called Forpheus. This machine, built by a Japanese company and now in its ninth generation, is equipped with state-of-the-art AI and multiple cameras that track and return shots with remarkable precision. With so many recent advances in AI, I became curious about how table tennis training could benefit from such progress. This article summarizes what I learned.

Training Technologies for Amateur Players

For amateur players, one of the simplest and most affordable tools is a cheap table tennis robot — something you can buy on eBay for around $200. It can shoot balls with various spins at the press of a button. I own one myself and have used it to train banana flicks and both forehand and backhand loops against backspin. It works reasonably well.

However, it has major limitations:

• Topspin balls are too slow, so it cannot mimic real match conditions.
• The robot is hard-coded: if you set it to one backspin level, it stays at that exact spin forever. No variation. In real matches, every backspin is different.
• You cannot learn spin-reading, one of the most important real-match skills. Reading spin requires watching the opponent’s racket at the moment of contact. A robot with no racket cannot teach this.

To compensate for the robot’s inability to produce realistic topspin balls for my loop training, I bought a rebound board — essentially a wooden plank covered in inverted rubber.

When you hit topspin onto it, the board returns topspin back at you, functioning like a human blocker. It costs under $100, but beginners may struggle. To use it consistently, you must hit the same spot with the same force, or the ball won’t come back to a hittable location. In my experience, you need to be at least USATT 1600 to benefit from it.

The Reality: The Best Training Still Requires Humans

In truth, the best way to improve table tennis skills is to practice with a partner or hire a coach. A human can vary spin, speed, placement, and timing in ways no affordable robot can. But partners require going to a club, and coaches in Maryland cost around $70 per hour.

Some clubs have advanced “robot coaches” that use real rackets. This does help you develop the ability to read spin from racket motion — but beyond that, the experience is not much different from a traditional ball-feeding robot. These machines are still hard-coded and cannot react dynamically the way human players do.

Where Current Technology Actually Stands

You’ve probably seen videos online showing humanoid robots rallying like professional players. Many of these are AI-generated fakes. One such video (shown above) replaced German national player Benedikt Duda with a humanoid robot — that footage simply isn’t real.

Not long ago, I attended a seminar where an MIT PhD candidate demonstrated a robotic arm completing a simple task: picking up a wine glass and moving it from one side of a table to the other without spilling. Watching that, it becomes clear how enormous the gap is between today’s robotics and the flashy AI-generated videos of robot table tennis “professionals.”

Actual humanoid robots do exist, but their table tennis level is roughly equivalent to a beginner — maybe USATT 600 at best. They can return basic topspin or no-spin balls, but offer no real threat.

The best real technology I’ve seen is the Japanese robot Miwa Harimoto tested. She said she was impressed by its consistency and had to work to return every ball. She even recommended adding more spin to make it a tougher opponent. After watching the full video, though, it’s clear she was playing at maybe 10% of her true power. If we mapped the robot to human levels, I would estimate its “rating” around USATT 1000 — still a meaningful technological achievement, but nowhere close to elite human play.

What Needs to Happen for Further Advancement?

I recently saw a clip from a Chinese robotics company demonstrating a humanoid robot walking very much like a real human. It became the top AI news story of the day because many viewers believed it had to be a real woman covered in robotic skin. That tells you exactly where the technology stands right now: walking like a human is still headline-worthy.

Before a robot can genuinely serve as a table tennis partner or coach, it must first master:

• walking naturally
• jumping
• bending
• rotating the torso
• generating fingertip-level force variations to grip a racket
• reacting in real time to unpredictable incoming spins

Only once these fundamental motor skills are perfected can true robot training partners emerge. After that, the remaining work will be fine-tuning details such as wrist torque, racket angle adjustments, and micro-timing.

My prediction? Maybe in 20 years we’ll see robots capable of genuinely challenging intermediate human players. What do you think? Feel free to leave a comment below.

If you’re interested in learning table tennis, check out my book From Blade to Brain: The Modern Table Tennis Intelligence on Amazon.

From Blade to Brain: The Modern Table Tennis Intelligence

All video clips were taken from the public domain on YouTube and transformed into GIF files by the author.

Advances of AI in Table Tennis Training was originally published in AI Advances on Medium, where people are continuing the conversation by highlighting and responding to this story.