Why we build it

Built for the nights
you practice alone.

Every hour of real practice has needed a second person — to rebound, to watch, to tell you the truth. We are engineers who play, and we are replacing that person with a machine.

Before writing a line of code, we stood on a court with a stopwatch. Out of one hour of solo shooting, roughly half was walking after the ball. That's the entire company, in one measurement.

— from the first page of our build log, still pinned above the bench

The Vision

A teammate on every court.

Coaching is rationed by geography and money. The players who improve fastest have a gym, a coach and someone to pass them the ball — not the most desire. We build the part of that you can carry.

Five kilograms folded. One person carries it out. Everything else in the design followed from that.

The Mission

Give the hour back.

One number decides everything we build: honest repetitions per hour. These three rules follow from it.

01

Reps per hour, or it's cut

A training tool is judged by one number: quality repetitions per hour. Every feature must raise it. Everything else — however clever — gets deleted.

02

Honest numbers only

The camera doesn't flatter you, and neither does this website. Release angle, arc, consistency — shown exactly as measured. Improvement starts from the truth.

03

Serviceable beats disposable

The feeder is a cartridge you can swap in seconds; wear parts unbolt instead of unglue. A machine you can fix is a machine you keep.

The Bench

Five schools, one bench.

This is a robotics company that builds for basketball. The people here come out of drone flight control, electric mobility, on-device perception and university robotics labs, and they brought their standards with them: a drive that repeats to the millimetre, perception with no cloud in the loop, a stop path measured in milliseconds.

USC

Elias Navarro

Founder · Computer Engineering

Timed his own practice hour and found most of it went to chasing the ball rather than shooting. Built the first arc-tracking prototype on a borrowed depth camera that month. Owns the shot model and the product roadmap.

UCLA

Kenji Sato

Drive & Structures

Drone propulsion and electric mobility before this. Owns the twin-roller feed: the spin-rate servo loop, the torque ripple that used to put a wobble on the ball, and the folding chassis that keeps the whole machine at five kilograms. Holds the feed to a repeatability figure, not a tolerance band.

MIT

Camila Duarte

Perception & AI

Runs the sensing stack: the solid-state LiDAR return path, its calibration against the 4K/120 stereo pair, and the fusion that turns both into a pose. Release angle, arc and joint tracking are all solved on the machine. No court has Wi-Fi, and no session leaves the unit.

Caltech

Amara Okafor

Controls & Safety

Real-time control and the whole interlock chain around it — motion planning, the watchdog, and the stop itself. The machine shares a floor with people who are not watching it, so the stop path is wired rather than scheduled: it holds even if the software is gone.

Berkeley

Marcus Bell

Sport Science

Former players and a strength coach with veto authority over every metric the product reports. They decide what counts as a rep, what counts as a clean release, and what a session summary is allowed to claim. A number that does not change your next session does not ship.

Every one of us plays. That is not a brand value — it is why the company exists.

Engineering, orders and support are answered by the people who designed the parts:sales@arc-x.ai

Train with it first.

First production run ships Q3 2027. Reservations are open now.

Reserve — $10