Edge Pong

Inspiration

What it does

How we built it

Inspiration

We wanted to make a game you can actually feel — not through a screen or a headset, but in your hand. Ping-pong was the perfect fit: fast, physical, all about timing. What if you could swing a real paddle at a ball that only exists on the wall, and feel the exact spot where it hit?

What it does

Edge Pong is a projected arcade ping-pong game. You swing a real paddle (or a mouse) at virtual balls, and when you connect, a smart paddle buzzes right where the ball struck — top-left, bottom-right, wherever. All the tracking, physics, and haptics run locally on the edge — no cloud, no headset. Miss your timing and the rally's over; land it and you score.

How we built it

Backend (Python): an authoritative game engine — fixed-step physics, swept collision so fast balls can't tunnel, a full rally system with serves and an AI opponent, scoring, and pose fusion that blends camera + IMU orientation using quaternions.
Frontend (Three.js + TypeScript): a 3D arena — table, net, paddle, opponent — driven over a WebSocket at 60 Hz, with sound and particles fired off the same impact events as the haptics.
Smart paddle (ESP32-C5): custom firmware streams motion over Wi-Fi and receives haptic commands, driving four vibration-motor quadrants through MOSFETs.
Haptics: four motor quadrants with bilinear interpolation, so the balance of the four motors encodes where on the paddle you hit.

Every physical piece also has a simulated stand-in — a fake camera and paddle feed the exact same pipeline — so the whole game was playable with a mouse long before the electronics were done.

Challenges we ran into

Honestly, most of the fight was hardware.

MOSFET wiring: our first motor spun the instant the battery connected, no code involved — a swapped Drain/Source, where the transistor's internal diode bypassed everything. Multimeter detective work to find it.
Power: the boost converter kept auto-shutting-off because the ESP32 didn't draw enough current. The fix — disabling Wi-Fi modem sleep — turned out to be exactly what we also needed for low-latency haptics.
Networking: with no home router, we ran the laptop as a hotspot. The paddle wouldn't connect until we forced it to 2.4 GHz; it kept landing on a flaky 5 GHz channel.
Feel: we measured the motor's stiction point (~18% duty), set the haptic floor just above it, and reworked the strength curve so even light hits land solid instead of fading to nothing.

What we learned

Your hand can't resolve 16 vibration zones on a paddle — 4 is plenty, and interpolation fills in the rest.
Quaternions beat pitch/yaw/roll for a swinging paddle: no gimbal lock, smooth blending.
"Fun over physically exact" — a slightly forgiving hit volume makes it feel responsive instead of broken.
Half of hardware is just measuring things with a multimeter before you trust them.

What's next

Wiring up the real camera + AprilTag tracking (the interfaces are already there), populating all the motors, and packing the electronics into the paddle handle for a proper self-contained controller.