Following the 2024 Paris Olympics, China’s tennis boom has sparked an explosion in AI robots. Shenzhen AceiiLab has launched the A1 robot, which moves along the baseline in a suitcase-style design to deliver serves. Using proprietary ultra-long baseline stereo vision and algorithms, it achieves precise trajectory prediction—without catching balls or resembling a human form—focusing solely on delivering realistic rally experiences through mobile ball feeding. Integrated with motion analysis, data tracking, and personalized training, the A1 builds a continuous service loop through a hardware entry point, redefining the consumer-grade AI tennis robot experience.

Article author, source: Hirdwire

During the 2024 Paris Olympics, on the afternoon Zheng Qinwen lifted her gold medal at Roland Garros, court bookings in China surged severalfold within a week. Two years later, the first to turn this surge into a business were a group of Shenzhen engineers who build machines.

Over the past year, at least four AI-powered ball robots developed by Chinese teams raised millions of dollars on Kickstarter. PongBot raised $2.65 million with an overfunding rate of 26,255%; SwitchBot, a leading smart home company, launched Acemate and earned $1.8 million in 30 days; Tenniix raised nearly $1 million in its debut campaign, after which its parent company, Yinghanxi Power, secured over 100 million RMB in three funding rounds co-led by Sequoia; the latest entrant, AceiiLab, is now delivering its first production units, with a retail price of approximately $2,000 per unit and an early bird price of $999.

But despite the hype, this sector hasn’t even clearly defined what it’s selling. Behind the label “AI tennis robot,” it could be a humanoid robot that plays tennis, a portable ball machine, or even a mobile app.

Why has the tennis robot赛道 suddenly exploded in popularity? What should a true AI tennis robot actually look like? Armed with these two questions, we met AceiiLab and its co-founder Chao Guang from Yisi Technology on a court in Shenzhen. We soon realized that in this competition, hardware is merely the entry ticket—the real key lies in whether you position this machine as a piece of equipment or as an opponent. Whoever can make users play longer and find it harder to leave will have the chance to define the term “AI tennis robot” first.

One: "A training partner who knows how to move into position has arrived on the court."

Before seeing the actual product, we imagined a tennis robot as a moving screen or a mechanical arm capable of swinging a racket.

When the AceiiLab A1 is rolled onto the court, it looks like a 20-inch suitcase—the lower half is the wheeled base, and the upper half is a detachable ball compartment that holds up to 120 balls. Previously, you had to carry the machine in one hand and drag the ball cart in the other; now, both components are combined into a single unit weighing approximately 20 kg, easily rolled on four wheels.

“We found that on the device, users’ core need isn’t for the machine to be light, but for themselves not to feel tired when moving it. Choosing a suitcase design eliminates the need for constant exertion, significantly reducing the sensation of fatigue,” said a representative from AceiiLab. “The suitcase format solves the mobility challenge while offering a user-friendly experience.”

After actually turning it on, we found that AceiiLab’s gameplay differs significantly from traditional ball machines—it’s a mobile, highly interactive product. “This might be the best AI tennis robot I’ve ever used; the realism of actual rallies enabled by its mobility is something almost no other product can match,” said a friend who participated in a live session playing against AceiiLab.

Place AceiiLab on the tennis court line, press the main unit’s power switch, and the side rail brackets will automatically unfold. After about ten seconds, the camera identifies the lines on the court, and the device begins patrolling along them. It doesn’t wander randomly across the entire court—only following key lines such as the center line, baseline, and boundary lines, like a disciplined player who never steps out of bounds.

Warm-up complete, the game is about to begin. Before kickoff, select your mode in the app—there are currently three main options: Combo & Movement, Free Practice, and Serve-and-Receive Training.

Taking the combination and movement mode as an example, the AceiiLab A1 will move along the baseline while delivering balls according to over 50 pre-programmed shot patterns—alternating between forehand and backhand, deep and short shots, and varying speed and spin—with a minimum interval of just half a second between balls.

Free practice returns the serving choice to the user, allowing them to select any mode they need—whether it’s cross-net rallies, same-side practice, or letting the ball move freely along the baseline. It can also deliver targeted shots or position itself along the sidelines to serve sharp-angled balls, transforming the entire baseline into a serving point, unlike traditional ball machines that repeatedly offer only a limited range of angles.

In the receive service practice mode, there is also a “ball boy mode”—you don’t need to press the start button; simply hold your racket upright, and the machine will automatically detect your ready state and serve the ball, as if a professional ball boy were standing across the net serving to you at all times.

The actual scene after the match begins is different from our imagined intense "rally." The truth about this "rally" is that when you hit the ball over the net, before it lands, the machine moves near the landing spot and immediately launches a new ball from that location, simulating that it has "returned your shot" by sending a ball back across the net. It never truly returns your original ball, but the back-and-forth rhythm still creates a genuine sense of rally.

However, this not only tests the user’s ability to adapt but also whether the machine can respond quickly and deliver precise shots. If there is noticeable lag, the rally experience will be significantly diminished. It took us at least dozens of minutes to gradually adjust to the rhythm of playing with the machine.

With 120 balls and a standard pace of play at a 3.0 level, you can expect to play for about 30 minutes. Since each ball is new, the overall rate of consumption will be relatively high. “Whether you’re a beginner or an experienced player, you’ll enjoy a unique experience.”

“The feeling of being unable to react” was a common first impression shared by multiple users we interviewed who trained with Aceii. Throughout the entire training process, we clearly sensed that this is not a product designed for absolute beginners with no prior experience. Its true strength lies in reinforcing forehand and backhand techniques, as well as footwork—comparable to the early-stage training method where a coach manually tosses balls for repetitive practice. It requires users to understand their own skill level and know how to configure serving parameters, making it a product that demands adjustment and adaptation between the user and the device. According to the developers, Aceii is designed to meet the needs of players at the 2.0 to 4.0 skill levels.

Two, it covers everything: coach, caddie, and bad friend.

There’s a hidden hierarchy in manufacturing: selling products is less prestigious than selling consumables, and selling consumables is less prestigious than selling services. For the past fifty years, tennis ball machine manufacturers have remained at the bottom of this chain, making only one-time sales that end once the product is sold.

Most table tennis robot manufacturers aim simply to sell their machines. But for others, a sale is the end point; for us, it’s just the beginning. According to AceiiLab’s leadership, robots often sit unused at home because they offer limited value—they can’t replace a coach—or because users lose interest after the initial novelty fades. AceiiLab aims to build a reverse flywheel: the more enjoyable the device, the more frequently users play; the more they play, the more they improve; and the more they improve, the more enjoyable it becomes. Hardware enables play; software keeps users engaged.

The first spin of the flywheel begins with a scorecard. After each round, you can view the training data: return success rate, highest streak, average ball speed, and maximum ball speed. There’s also a landing map that plots every return—yours and the machine’s—onto the court, clearly showing which shots landed in or out.

More importantly, it also teaches you. The machine records your swing video and analyzes it across five dimensions—clubface angle, stability, timing, power chain, and setup—then tells you how this round performed and compares it to your past data.

After you submit your data for AI analysis, the system will identify weaknesses, providing not only the reasons and recommendations but also recommending relevant video courses to help you improve your situation.

In AceiiLab’s vision, a day with the AI coach would look like this: Yesterday, your backhand follow-through was insufficient and your hip rotation lacked power—today, you focus on precisely those two areas, practicing shadow swings while comparing them to standard form, and then the machine tests you again. The entire coaching process has been faithfully replicated within the AceiiLab system.

Moreover, beyond practice, it aims to add more fun. While playing, you can activate the ball boy mode—wave your hand and the ball is served—or switch to game mode to complete levels, turning mundane training into an engaging challenge. If solo practice feels dull, you can share exciting moments with the community and easily invite friends to join you. “The online world has been fully optimized by algorithms; the only variables remain in the physical world. A tennis robot sprinting across the court naturally draws attention,” said a related spokesperson.

Many of these design choices stem from direct feedback from the market. During a road test in North America, a user spent half an hour playing and asked a question the team hadn’t anticipated: “Am I never going to win against it?” This made them realize that while the machine can play indefinitely, humans cannot. People need levels, wins and losses, and an opponent they can actually beat.

Thus, challenges were introduced, and even rules for machine defeat: if the ball lands more than 50 centimeters away, the machine loses. Because a machine that has the option to lose is more like a teammate than one that is always right.

All this data, accumulated over time, becomes the reason users won’t leave the machine. “When the machine holds years of your progress data—and even the training data of your best friend—even if a competitor is $500 cheaper, you won’t switch,” said a related executive. When parameters can be matched and molds can be copied, the time users have invested is the one thing no one can take away.

We also see that, in the future, the competitive advantage of hardware companies may lie not in product specifications, but in transforming their brand into meaningful user relationships. AceiiLab is actively delivering continuous services through AI and data within a specific use case.

Three: A New Species Emerges on the Most Competitive Supply Chain

Why does a tennis robot only run along the baseline? Why doesn’t it return the ball? These may seem like technical issues, but in the early stages, they are actually product decisions. To understand these choices, you first need to understand the people making them.

AceiiLab is backed by Yisilab, a team assembled by veterans of the service robotics industry. Founder Liu Liqian worked as a structural engineer at DJI, a warehouse robotics engineer at Blue Dolphin, and a systems engineer at KUKA, before shifting focus to commercial cleaning and lawn-mowing robots. He served as CEO of Orca, a lawn-mowing robot company, and led its campaign to become the top-funded overseas crowdfunding project that year. This team of over thirty members has spent many years gaining hands-on experience in the service robotics industry.

Because they’ve been at it for so long, they realized that whether it’s a service robot or a tennis robot, both operate within a set of specific rules. Lawn mowers recognize the boundaries of a lawn; tennis robots recognize the lines on a court—both are governed by rules, with standardized courts and ball trajectories following parabolic paths. Tennis is nearly the most rule-bound sport in the real world, which was one of the reasons they were early believers in this direction.

But there is a fundamental difference between the two types of machines. With a lawn mower robot, there is an employer-employee relationship: the user only cares that the grass is cut cleanly, not how the robot does it. It serves the outcome, and its value ceiling is at most the wage of a hired worker. A tennis robot, by contrast, serves the process. Players care about the speed of each shot, its landing point, and whether the form is correct. The robot perceives and responds to you every second—this is a partnership. What they ultimately aim to achieve is enabling users to train and compete against high-level coaches and partners anytime, anywhere.

Once they’ve figured out what they want to do, their next question is how to do it—maximizing value within a price range that casual players can afford.

First, they must overcome the cost barrier. The team’s approach is to build on established industries: the robot’s chassis directly uses a decade-refined mobility solution from vacuum cleaners, the wheels are taken directly from suitcases, and the movement system employs a solution similar to lane-keeping assistance.

The money saved was reinvested in time and design flexibility. But saving money doesn’t mean cutting corners—where it mattered most, they chose to build it themselves, such as the most expensive component of the machine: the robot’s “eyes.”

Off-the-shelf stereo camera modules have too short a baseline—they can only determine where the ball lands, not how it travels through the air. Without clearly tracking its trajectory, the system cannot anticipate positioning, making any sense of “feel” impossible. To solve this, the team developed a custom ultra-long-baseline stereo camera that calculates the ball’s first landing point within 10 milliseconds of launch, continuously observing and adjusting its prediction throughout flight, and fully mapping the entire trajectory within 50 milliseconds. Interference from outdoor lighting—such as backlighting, glare, or arena lights—is minimized, ensuring accurate ball detection and judgment under all lighting conditions.

Another major focus is on the algorithm team, which is the largest team in the company, primarily responsible for enhancing motion algorithm analysis capabilities. They are already working on multimodal body and movement analysis. For opponent simulation, current user modeling remains primarily rule-based—extracting coaches’ expertise into six parameters such as footwork and gestures, and encoding them into rules. This approach already enables training simulations of specific playing styles, such as “Nadal’s left-handed topspin shot.”

Eyes and algorithms are the two areas where they’ve spent the most money on this machine. But what truly defines the machine is where they chose not to act. After all, adding features only requires money; subtracting them requires judgment—and this team’s judgment on AI tennis is written on a list of “don’ts.”

IV. A "Do Not" Checklist: Understanding the Trade-offs of Consumer-Grade AI Tennis Robots

In today’s landscape of diverse product forms, we see Aceii choosing many things not to do: not deploying robots that roam the entire court, but only moving along the baseline; not adding flashy features, instead crafting the experience around mobile serving; not attempting to fulfill every requirement at once, but first focusing on meeting the needs of advanced players at the 2.0 to 4.0 level. Each of these “not doing” choices is grounded in clear reasoning.

Among these, what intrigued us most was why not build a humanoid robot to play tennis? Zhao Guang said: “Humanoid robots playing tennis test the capabilities of robots; we are building tennis robots that deliver the experience of playing tennis—products that consumers will genuinely want to buy. On a flat court, wheels are more efficient than two legs, just as using a washing machine is more efficient than hand-washing clothes.”

Behind the decision not to return shots lies a training philosophy. In Aceii’s view, serving and returning are two separate actions—the serve sets the rhythm of training. When the machine alters its own rhythm in pursuit of returns, the overall training value diminishes. Moreover, if the return success rate is too low, the dynamic reverses: instead of the machine training the player, the player ends up leading the machine around.

The reason for choosing the baseline as the movement focus is driven by market demand. According to industry statistics, 70% to 80% of target users' shots are baseline rallies. By holding the baseline, you secure 70% to 80% of the user experience.

So why, if this machine is restrained in every other aspect, does it spare no cost when it comes to "mobility"?

To them, mobility is the dividing line between AceiiLab and traditional ball machines. A stationary machine interacts with a person only once in the morning—turn it on, set it up, then go your separate ways. But a mobile machine interacts once per second: you hit left, it chases left; you hit right, it chases right; it sees you, and responds. This also creates deeper interaction. Moreover, mobility brings a sense of opponent, enables multi-point feeding, net rallies, ball-boy-style play, and fosters a subtle emotional connection. At the event, we also observed early users praising the “duel” experience enabled by mobility: “Being able to receive serves while moving creates a much stronger sense of immersion and realism.”

When you reverse these “don’ts,” AceiiLab’s path becomes clear. Chinese hardware companies have followed the same path for thirty years: making a product category cheaper, only to be outdone by newcomers who make it even cheaper—launching one product, fighting one battle, with profit margins thinning with each generation until there’s no battle left to fight. AceiiLab aims for a different route: first selling the machine by delivering a completely reimagined serving experience, then reducing costs through mass production, retaining users through mobility and interaction, and finally turning accumulated data into services.

On this path, restraint is not conservatism, but patience. It’s still too early to conclude whether this path will succeed. The first batch of mass-produced units only began delivery in July this year; renewal rates for subscriptions, course completion rates, and community engagement levels all await real-user data.

Time will tell—which side the users on the field will choose.