Industrial Design and Mechanical Engineering student driven by a passion for designing products that combine thoughtful aesthetics with practical function.
Drawing on mechanical intuition, manufacturing knowledge, and user-centred design principles, he seeks to create products that work as well as they look.
- +61 416 984 667
- naruncheng@gmail.com
- Melbourne, Australia
Personal Project
AI Tennis Trainer
2026
- — Engineering
- — Personal Project
- — Concept to Prototype
The AI Tennis Machine is an ongoing personal project to build an AI-powered tennis training machine capable of launching balls that mimic a variety of real shots.
This is a full-stack project spanning hardware and software, built around three core systems: a launch system that fires balls with varied speed, spin and placement; an AI and vision control system that tracks player and ball movement so the machine can respond intelligently during training; and a power system to run it all on court.
Unlike a traditional ball machine that feeds balls to fixed locations, the AI Tennis Machine supports movement-based drills that train positioning, recovery and court coverage.
Working CAD assembly of the tennis ball launch system. Two sets of flywheels are mounted inside a sheet metal frame alongside the feed tube. Each flywheel is driven by a 3420 motor. Varying the speed difference between the top and bottom flywheels allows control over topspin/backspin. The working design utilises a direct drive setup, with a jaw coupling between the motor shaft and the flywheel axle to absorb vibration when a ball is fed.
Additional views of the launch system CAD. Some quick calculations show that given motor speed at a peak of 7000RPM, the tangential velocity at the point of contact on the flywheels reaches up to ~145km/h. Given average tennis pro tour forehand speeds reach around ~120km/h, this drivetrain setup should be sufficient for a first prototype.
Some of the components for the launch system drivetrain initially purchased.
A sheet metal subframe connects to the launch system subassembly via two pairs of bearings which allows pitch and yaw rotations. Adjusting the pitch gives control over shot height, and adjusting the yaw gives control over left and right shot placement. These two rotation axes will likely be driven by lead screw motors with simple linkage mechanisms.
This entire assembly is then mounted inside a compact frame built from 20 series aluminium extrusions. The ball feeding mechanism sits atop the launch system, which utilises a carousel mechanism to funnel tennis balls into the feed tube below.
Autodesk Fusion renders of the working CAD. Several aspects of the hardware are still being worked on, including the control electronics and the onboard power supply, as well as the compute and camera hardware for the AI vision system.
The Nvidia Jetson Xavier NX which provides the edge compute for the Ultralytics YOLO vision pipelines. Edge compute was chosen over a cloud solution, since a cloud approach would require a reliable internet connection on court and introduce network latency, which is unacceptable for real-time ball and player tracking.
The AI vision system will primarily be used for two key functions.
The first is ball feeding. One issue with existing tennis ball machines is that ball feeding is controlled either via a mobile phone app or a remote controller. While playing, carrying a phone or even a small controller can be cumbersome, especially if the player needs to quickly readjust their hands from the phone/controller to the racquet.
An AI vision pipeline could therefore read the body language of the player from the other side of the court. For example, the player could raise their right hand up to request a forehand topspin shot from the machine, reach out wide for a forehand dropshot, and vice versa with their left hand. While the exact mapping of player action to machine response can be considered and iterated further, this concept would provide a far more streamlined and ergonomic process compared to using a remote.
The second function of the AI vision system is player-specific training. By tracking player and ball movement over time, the system can identify each player’s strengths and weaknesses across sessions, such as a weaker backhand, slow recovery to the baseline, or poor coverage on wide balls. The machine can then tailor future sessions to the individual, targeting areas of weakness while reinforcing strengths, so training adapts as the player improves.
AI vision progress updates coming soon.
- +61 416 984 667
- naruncheng@gmail.com
- Melbourne, Australia
Industrial Design and Mechanical Engineering student driven by a passion for designing products that combine thoughtful aesthetics with practical function. Drawing on mechanical intuition, manufacturing knowledge, and user-centred design principles, he seeks to create products that work as well as they look.