RallAIMate machine concept render — dark tower machine with blue LED accents and a hopper of pickleballs
Open hardware · In active development
Product design concept · final hardware may differ

RallAIMate

The pickleball training machine that's actually yours. Open API. Local-first. No subscription, ever.

Open by design

Every part. Out in the open.

Commercial machines hide a cheap board behind a locked app. We pull the whole thing apart and rebuild the control stack where you can see it.

Scroll through the build. Concept visualization · scroll or choose a component below.

Yours to run

Yours to modify. Yours to keep.

An ESP32 brain, a REST + WebSocket API, and drills that are just JSON files. It works in a garage with no internet at all.

Smart-machine training. Dumb-machine price.

Built in public, phase by phase. Follow the teardown, the firmware, and the first serve.

See the Machine Explore the Technology
~$795target build cost, all-in
5 chindependent motor channels
$0/mosubscription. Forever.
100%open API & local-first
Scroll to disassemble
The Machine

Meet the concept.

Dark chassis, blue LED accent lines, an open hopper, and dual-wheel drive — the long-term design target for the retrofitted machine. Full concept sheet below, honest Phase 1 numbers beneath it.

RallAIMate concept sheet — machine renders from front, side, rear, and top with feature callouts: AI-powered intelligence, dual motors, app control, 120-ball capacity, spin control, elevation range, and battery life

Concept target vs. Phase 2 hardware — the honest sheet.

The render above is where the platform is headed. Day one runs on a stock donor machine with an ESP32 swapped in — and we publish the real numbers, because that's the whole point of an open build.

Spec Concept target Phase 2 (stock donor + ESP32)
Ball capacity 120 balls 50 balls — the donor's physical hopper
Ball speed up to ~75 mph ~20–37 mph on stock motors; brushless upgrade is a later stretch
Spin control ±9 continuous levels pending teardown — depends on whether the wheels are independently driven
Elevation 20–60°, motorized 20–60°, motorized — confirmed donor spec, real on day one
Battery life 6–8 hrs 1–2 hrs — stock pack, per the listing
Mobility all-terrain wheels stationary — smooth surfaces; Mecanum base is Phase 5
App, stats & AI coaching full app suite raw REST + WebSocket control — web client is Phase 4, vision is Phase 1

Phase 2's win condition, stated plainly: build our own brain for a dumb machine and prove full programmatic control. The roadmap tracks the remaining concept features — and every phase ships its real numbers.

01 · Start here: Vision

First, prove players can see their progress.

You put in the reps. See what is improving. RallAIMate Vision is a proposed camera-based training tool that turns practice into a map of where your shots land. Start with a phone, prove the feedback is useful, then offer a dedicated camera puck, the open training machine and a complete training bay.

RallAIMate Vision

A fixed camera. Every bounce, mapped to the court.

The prototype uses a fixed camera to identify the ball and map detected bounces onto the court. The first goal is reliable placement feedback and a shot log. Speed estimates follow only after validation. Local processing and optional sharing are design goals.

  • No waiting. The donor machine takes 4–6 weeks to arrive. Vision v0 is software on a phone camera. Work starts now.
  • Bigger market than machines. Players and coaches with existing machines are the first audience to test.
  • It earns the name. Useful placement feedback is the first proof point. Accuracy comes before additional features.
  • A learning advantage. Players keep their sessions locally. Only recordings explicitly shared with permission can support a future training dataset.
vision — session review (illustrative)
Illustrative Vision session: court placement map with 50 example shots, 32 in the target zone, and a next-drill suggestion

Illustrative interface using synthetic data. This explains the intended experience; it is not a measured result or a working product screenshot.

Court Pod · concept render

One pod on one net post. Two lenses, one per half.

Two global-shutter cameras in one body: placement, in/out and kitchen faults on both halves from a single clamp-on device. Scroll to take it apart, explore the components, and bring it back together.

RallAIMate Court Pod concept render: graphite body, cobalt band with wordmark, two hooded lenses, quick-release base
Court Pod · inside the concept

Two lenses. One pod.

Concept render. Lens geometry is still being resolved: on the real unit the two lenses spread 90° apart so each faces one half of the court.

Version Form When Hardware
v0 Phone or webcam on a tripod, browser app Weeks 1–4 $0
v1 puck Raspberry Pi 5 + Global Shutter camera, mount for supported fixed positions, Wi‑Fi to the Studio app Weeks 4–8 ≈ $265–305
v2 Compute-module redesign, molded shell. Production BOM target ≤ $150 (assumption) Pilot / launch retail $299–349 target

Vision opens the door. The machine expands it. Win on useful feedback first, then ask players to upgrade the hardware around it.

02B · Mobile & Apple Watch

Bring your game into practice.

Your missed backhand reset. Your drop that lands too deep. The return that comes up short. Recreate My Game turns those recurring weaknesses into drills you can repeat in a Training Bay or on court.

Build a personal game profile from player observations, coach notes and supported Vision sessions. We use “game handicap” to mean the weaknesses holding a player back; this is a training profile, not an official handicap or competitive rating.

Mobile App

Understand, recreate, improve.

  • Your game profile. Choose the situations that cost you points. Add coach feedback and review supported session data before accepting a suggested focus.
  • Recreate My Game. Convert a selected weakness into targets, repetitions, feed direction and difficulty. Preview the setup and adjust it to your space and equipment.
  • Practice with a purpose. Run the saved drill using manual feeding, a compatible machine or the Training Bay. Each drill explains which game situation it is intended to address.
  • Compare like with like. Compare target success and consistency across the same drill and conditions. Screen-impact scores stay separate from real-court bounce results.
mobile app — game profile & drill setup (concept)
RallAIMate mobile app concept showing a game profile with backhand reset, third-shot drop and deep-return weaknesses, beside a personalized drill setup

AI-generated mobile app concept. All screens, examples and improvement messages are illustrative; no app or training outcome is represented as delivered.

Apple Watch Companion

Keep the session within reach.

Planned role: session controls, glanceable progress and vibration cues. Shot results come from Vision and the connected session; the watch does not independently track ball placement.

  • Start, pause, resume and end, with a readiness confirmation before machine feeds begin
  • Vibration cues mark set changes and rest intervals
  • A short summary points back to the phone for detailed review
  • The machine’s physical emergency stop remains independent of the app
apple watch — drill in progress (concept)
RallAIMate Apple Watch app concept with a backhand reset drill, 12 of 30 balls, a pause control and a completed-session summary

AI-generated Apple Watch companion concept. Test connection loss, delayed commands and supported device combinations before release.

Scope note: the current Studio budget covers only the browser-based personalized-drill prototype. Native mobile and Apple Watch apps are a later release with a separate engineering estimate before a date is committed.

Training Library

Every drill, on video.
Machine settings included.

Each session ships as a video walkthrough plus the exact drill file that produced it. Watch the drill, tap once, and your machine runs the identical feed — same speed, same spin, same placement.

Featured Session

The 100-Ball Third-Shot Drop Ladder

Alternating cross-court targets, speed stepping up every 20 balls, topspin bias increasing each round. The drill file that runs it is 14 lines of JSON.
Dinks12:40

Kitchen-Line Dink Triangle

Three rotating soft targets at the NVZ line. Consistency under fatigue.

Beginner+90 ballsdrill file incl.
35 mph
Drives9:15

Drive Defense at Full Pace

Body-targeted drives at ascending speed. Block, reset, counter.

Advanced120 ballsdrill file incl.
+/- 1400 rpm
Spin14:05

Reading Heavy Topspin & Slice

Alternating spin profiles from independent dual wheels. Learn the flight cues.

Intermediate100 ballsdrill file incl.
Footwork11:30

Lateral Shuffle & Recover

Random left-center-right feeds force honest court coverage.

All levels80 ballsdrill file incl.
transition zone
Transition16:20

Mid-Court Reset Gauntlet

Survive the no-man's-land: low balls at your feet while you advance.

Intermediate+110 ballsdrill file incl.
ABC
Precision13:50

Baseline Target Boxes

Deep returns into called zones. Session heatmap scores every ball.

All levels100 ballsdrill file incl.

The library grows with the build — every prototype session is being recorded from day one, doubling as the training dataset for the computer-vision phase. Follow the build to catch each drop.

The Technology

Serious engineering, zero black boxes.

Commercial "smart" machines hide a cheap microcontroller behind a locked app. RallAIMate rips that out and rebuilds the entire control stack in the open — every motor, every message, every line of firmware.

The Brain

An ESP32 runs the whole machine.

The stock control board comes out. In goes a dual-core ESP32 driving five independent channels — with hard-realtime safety where it counts.

  • Two throw wheels with independent speed control — the speed differential is the spin control
  • Ball feed, horizontal oscillation, and vertical elevation on separate channels
  • Hardware e-stop on an interrupt line — sub-millisecond cutoff, no software in the loop
  • Over-the-air firmware updates: flash new behavior from the couch, not the bench
channel map — firmware v0.x
CONTROL ARCHITECTURE · CONCEPTESP32Wi-Fi + BluetoothLOCAL CONTROLThrow wheelsMOTOR CONTROLBall feedMOTOR CONTROLPan + tiltMOTOR CONTROLSafety inputE-STOPOne controller. Five motor channels.
The API

Every drill is an open JSON file.

REST for commands, WebSocket for live telemetry. If you can write JSON, you can program the machine — from a browser, a script, a Raspberry Pi, or anything else on your network. No cloud round-trip. No account. It works in a garage with no internet at all.

  • Drills are portable files — share them, version them, generate them programmatically
  • Live telemetry stream: wheel RPM, feed count, battery, machine state
  • Local-first by design: your network, your data, your machine
third-shot-drop-ladder.drill.json
// one step of a drill — that's the whole schema
{
  "position":    "cross-court-left",
  "speed":       34,          // mph
  "spinTop":     1.00,        // wheel A throttle
  "spinBottom":  0.62,        // wheel B → topspin
  "elevation":   12,          // degrees
  "intervalSec": 2.5
}

# run it from anywhere on your network
curl -X POST http://rallaimate.local/drill \
     -d @third-shot-drop-ladder.drill.json
Spin Physics

Spin is a number, not a knob.

Two counter-rotating wheels grip the ball. Run them at the same speed and the ball flies flat. Run the top wheel faster and you get topspin; bottom faster, backspin. If the donor's wheels are independently driven — the first thing the teardown will confirm — spin becomes a continuous, scriptable variable, and drills can morph spin ball-by-ball.

  • Continuous spin spectrum, not 3 preset modes
  • Ramp spin mid-drill to train reading ball flight
  • Honest engineering: ball speed caps on ball physics, not marketing horsepower
dual-wheel differential
DUAL-WHEEL DRIVE · SIDE VIEWTOP WHEEL4,200 rpmBOTTOM WHEEL2,600 rpmIllustrative wheel speeds · not measured performance
Spin
Angle
Speed
Mobility · Roadmap

A machine that takes its position.

The mobility base rides on Mecanum wheels — four independently driven rollers that let the machine strafe sideways, glide diagonally, or rotate in place. Between drill steps it repositions itself to preset court spots, so one drill can feed from the left sideline, the center, and the right without you touching it.

  • Omnidirectional Mecanum drive — built for smooth court and garage surfaces
  • Preset positions callable from the same drill JSON
  • Low-slung battery ballast counters launcher recoil
mecanum kinematics
MECANUM BASE · PLANNEDNETRandom movement + ball feeds · concept demo
Moving to position
Why Open

Own the machine. Own the data.
Own the roadmap.

Every other machine on the market is a rental of its own features. RallAIMate is built on three non-negotiables.

Open API, open firmware

The full REST/WebSocket spec and firmware are yours to read, fork, and extend. Build your own client. Script drills in Python. Wire it into anything.

Local-first, no lock-in

The machine runs on your WiFi — or its own access point with no internet at all. No account, no cloud dependency, no feature paywalled next quarter.

Hackable by design

Documented wiring, 3D-printable parts, a published BOM. Upgrade motors, add sensors, port it to a tennis or table-tennis donor — the platform is the product.

The Market Gap

Smart-machine training. Dumb-machine price. Open everything.

The market splits into cheap launchers that just spit balls, and $1,900+ app machines that rent you your own drills. Nothing is open — at any price.

Basic launchers "Smart" machines RallAIMate
Typical price $300–900 $1,900–3,900 ~$795 target
Programmable drills ✕ remote only ✓ via locked app ✓ open JSON files
Independent spin control ✕ or presets ✓ ✓ continuous, scriptable
Open API ✕ ✕ closed ecosystem ✓ REST + WebSocket
Works without cloud ✓ (no smarts) ✕ app + account ✓ fully local
Subscription — often, for full features never
User-repairable / moddable ✕ ✕ warranty-voiding ✓ that's the point
Build Roadmap

Built in public, phase by phase.

Every phase is documented — wiring photos, firmware commits, drill session videos, and the mistakes included.

PHASE 1 In Progress

Vision v0 → v1 Puck

Starts now on a phone camera (weeks 1–4), then a Pi 5 Global Shutter puck (weeks 4–8). Placement heatmaps first; speed only after validation. Processed locally.

PHASE 2 Next

Teardown & ESP32 Swap

Gut the donor machine's stock board, wire in the ESP32 brain, bench-test all five motor channels with a hardware e-stop.

PHASE 3 Next

Open Firmware & API

REST + WebSocket control layer, drill JSON schema, OTA updates. Machine-agnostic — the same firmware will drive future sports.

PHASE 4 Next

Web Drill Designer

Court-diagram UI: drop pins, sequence shots, save and share drills. Live control panel with manual override.

PHASE 5 Roadmap

Mecanum Mobility Base

Omnidirectional drive: the machine repositions itself to preset court spots between drill steps.

PHASE 6 Roadmap

Multi-Sport Platform

Port the firmware to a table-tennis robot donor — same API, different motor scaling. One platform, every racquet sport.

Follow the Build

Be there when it serves its first ball.

Get the build log, new training videos, firmware releases, and the open drill library as they drop. No spam — just the machine coming to life.

Open hardware project · in active development · documented end to end