Stewart platform machine keeps a ping pong ball bouncing

Nov 11, 2026 Last reply: 1 day ago 3 Replies
Jump to replies (3)

The toughest part of this build had to be the arms themselves. Early versions of the linkage design ran into joint limits and self-collisions in simulation, so extra degrees of freedom had to be worked in before anything could move properly. Getting past that hurdle is what makes the finished machine so satisfying to watch: a hexapod on a Stewart platform, balancing and juggling a table tennis ball on a paddle, with the ball kept bouncing continuously through control of its position.



The construction is where the craft shows. Every part was modeled in Fusion 360 across three project files covering the six arms, the base, and the central triangular end-effector, then cut from 3 mm 6061 aluminum on a desktop CNC mill driven by Universal G-code Sender. Each multi-joint arm goes together with bolts, nuts, and ten bearings, which sounds like a fiddly assembly repeated six times over.



The motion side is handled by six NEMA 23 steppers through 10:1 planetary gearboxes on StepperOnline CL57T drivers, all fed from a 36V 14A supply, with a Teensy 4.0 running the show. The stepper movements come across as smooth. For sensing, two See3CAM_24CUG cameras track the ball, and the machine computes its 3D position from that image data at 120 FPS, while the tracking views themselves report frame rates above 400 FPS. A Windows machine with a GPU runs a Unity application that mirrors everything on a connected screen as a real-time 3D simulation alongside the stereoscopic tracking views.



What stands out most is the math made visible: handwritten derivations in a notebook flow straight into C# scripting, and a 3D animation explains how motor angles come from intersecting circular and spherical paths of the linkage joints. The open question for me is endurance. The machine juggles steadily on camera, but nothing stated tells us how long a run lasts before the ball drifts off the paddle, and that seems like the next thing worth measuring.



Project by Electron Dust. Watch it on their YouTube channel:

formatting link


The choice of 10:1 planetary gearboxes on those NEMA 23 steppers looks clever to me. A parallel kinematic machine lives or dies on stiffness and resolution at the platform, and the reduction should help with both, at the cost of some top-end arm speed. Combined with the closed-loop CL57T drivers, the smooth motion reported makes sense as a result. To me that gearing seems like the right call for a bouncing task.

How are the two See3CAM_24CUG cameras positioned relative to the paddle and the platform? With stereoscopic tracking feeding the 3D position at 120 FPS, I am curious whether the pair sits wide apart for depth accuracy or closer together to stay out of the arm workspace. Also, does the calibration between the camera frame and the platform frame happen once at setup, or does it get refreshed between runs?

One variation worth trying: put the handwritten derivations straight into the Unity app as a live visualization, so the intersecting circular and spherical joint paths update in real time next to the simulation instead of only appearing in the explanatory animation. Since the C# code and Fusion 360 files are already on Github, someone else could add that overlay too, and it would make the kinematics easier to sanity-check when tuning the bounce.

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required