This Year's Theme: Physical AI

GridMind Robotics Competition

Most AI you've used lives inside a screen — a chatbot, a game, a recommendation. Physical AI is different: a camera SEEs the real world, code THINKs about what it sees, and then it actually DOES something about it. GridMind is your team's own Physical AI project — your KRIA board's camera is its eyes, your code is its brain, and the moves it reports back to the referee are its hands.

See the Full Schedule

See, Think, Act

See Your KRIA board's camera scans the grid
Think Your vision model identifies each object
Decide Your code claims a match, declines, or asks for a hint
Act The referee flips the physical cards and reports the result

A matched pair, mid-game — most of the grid stays face-down until a team's board chooses to reveal it.

What is the Robotics Competition?

This year's bootcamp finale replaces the hackathon with GridMind, a head-to-head robotics competition. Two teams share a grid of face-down cards, each hiding an object that appears on exactly two cards — a classic memory-match game, except each team's KRIA board plays it with its own camera and its own vision model. There is no shared "referee vision system" watching the grid: when a card is revealed, both teams get to look at it and remember what they saw, but each team's board decides for itself, on its own turn, whether two cards match.

Bootcamp Robotics Competition Details

Teams spend the bootcamp building an autonomous player for GridMind: eyes (a vision pipeline that recognizes objects on the grid), a brain (a decision loop that picks pairs, claims matches, and manages hints), all running on the team's own KRIA board, with no human input once the match clock starts. On competition day, every match is played and scored automatically by the GridMind referee software; there's no judging panel — the scoreboard decides the winner.

In the Robotics Competition, participants will learn:

  • Physical AI: Connect a camera, a model, and a decision loop into one system that sees, thinks, and acts in the real world — not just on a screen.
  • Computer Vision: Build and tune a vision pipeline that has to work reliably under real competition conditions, not just a demo.
  • Autonomous Decision-Making: Design a turn-taking strategy — when to claim a match, when to decline, when to spend a hint — with no human in the loop.
  • Networked Protocols: Talk to a live referee over the wire protocol, handling turns, timeouts, and scoring updates in real time.
  • Collaborative Learning: Engage in teamwork to apply new skills effectively under a real deadline.
  • Competitive Strategy: Every outcome is scored — a wrong guess costs points, so knowing when not to claim a match matters as much as recognizing one.

The Rules

Format
  • Six teams, two pools of three, round robin (three matches per pool).
  • 6x5 grid: 30 face-down cards, 15 objects, each appearing exactly twice.
  • Best record per pool advances to the Grand Final.
  • No elimination — a bad match doesn't knock you out of your pool.
  • Tiebreakers: most wins, then most total pairs matched.
Scoring
  • Correct match: streak bonus (1/2/3/4 pts) + speed bonus (+2 within 40s, 0 within 80s, −2 within 120s).
  • Wrong claim: flat −2, turn ends immediately.
  • Correct decline: zero, always.
  • Timeout: flat −3, always.
  • Paid hint: −1 each time.
Hints
  • Paid — for a specific object, mid-match. Capped at 2 per match, only while your score is positive.
  • Free — one shared hint, delivered during the ~60-second pre-game window only. Can't be requested again once the match starts.

Every team builds their own player for this competition — see How to Build Your Own GridMind Client for the exact wire messages and required integration points behind every step above.