{"slug": "pro-wrestling-for-reinforcement-learning-ducks", "title": "Pro wrestling for reinforcement-learning ducks", "summary": "Pollen Robotics is hosting QuackDown, a livestreamed simulator sumo wrestling tournament for reinforcement-learning policies controlling its 25 cm open-source MicroDuck biped robot, with the winner taking the Golden Beak Belt. The event, scheduled for October 2, 2026, features eight teams competing in best-of-3 matches in the MicroDuck simulator, with a side-card Stanford vs. Berkeley waddle-off, and is open for sponsorship and team registration.", "body_md": "### Meet\n\nA 25 cm open-source duck family. Looks harmless. Waddles.\n\nOctober 2, 2026 / livestreamed simulator sumo\n\nPro wrestling for reinforcement-learning ducks.\n\nFirst rule: you don't quack about it.\n\nStreaming live on YouTubeChannel TBA\n\nThe fighters\n\nEvery fighter starts as the same 25 cm open-source duck. What turns it into a contender is the policy — born cute, trained in the sim, sent to sumo for the belt.\n\nReinforcement learning drills ring control, pushing, dodging, recovery. [Clip via @jpgallegoar on X →](https://x.com/jpgallegoar/status/2095086006307266716)\n\nBest of 3 in the simulator. Last duck standing takes the Golden Beak Belt.\n\nThe bracket\n\nA tighter card keeps the stream moving: four quarterfinals, two semifinals, and one final. Every team starts on equal footing.\n\nRules of the sim\n\nThe ring is a plain circle with a hard, high-contrast edge. Same spawn spots and the same physics every match, plus room on the boards for sponsors.\n\nEvery fight runs in the MicroDuck simulator, not with physical robots.\n\nEach bout is best of 3 rounds — first policy to win 2 takes it; a 1:1 split goes to a decider.\n\nA round is won when the opposing policy drives, trips, or gets pushed out of the circle.\n\nSubmitted policies must run at event settings and use the approved control interface.\n\nReferee reset only for simulator faults, deadlocks, or unrecoverable spawn states.\n\nTraining camp\n\nTeams write the controller that decides how their MicroDuck moves, balances, commits, feints, and recovers inside the simulator.\n\nReinforcement learning turns scrimmage time into pushing, dodging, edge awareness, and last-second save behaviors.\n\nFinal policies are submitted before QuackDown so the livestream can run fair, reproducible matches.\n\nThe title\n\nThe winner gets the Golden Beak Belt: a ridiculous, glorious simulator-wrestling prize for the policy that owns the ring.\n\nAwarded on stream after the final best of 3. The belt belongs to the team, but the bragging rights belong to the policy.\n\nSchedule\n\nThe 8-team roster goes live with shields, walkout names, and policy teasers.\n\nTeams test movement, pushing, recovery, and ring awareness in public scrims.\n\nFinal RL policies are locked for inspection, reproducibility, and match setup.\n\nEvery first-stage matchup and the QuackDown path are revealed on stream.\n\nStanford vs. Berkeley runs as a side-card simulator waddle-off.\n\nStreams live on YouTube — the broadcast crowns the first QuackDown Champion and awards the Golden Beak Belt.\n\nEnter the bracket\n\nEight teams make the QuackDown card. Tell us who's fighting and how to reach you — we'll send simulator access, the policy submission window, and your bracket seed.\n\nSponsor the spectacle\n\nOwn a spot inside the QuackDown arena — the mat, the apron, or the back wall that renames the whole event. Spin the arena and pick a position.\n\nDrag to spin · hover a slot · pick a position below\n\n01 · Ringside boards\n\nA ring of sponsor boards wrapping the whole fight floor — sell them as a set or one by one.\n\nRingside boards · 12 of 12 open\n\nDuckFightClub by you — your name on the back wall, the stream slate, and the belt ceremony.\n\nOwn one match card, lower third, replay frame, and winner graphic.\n\nRotating placements for intermissions, scrimmage clips, and policy recaps.\n\nStart a chapter\n\nDuckFightClub is a format anyone can run. Bring the ducks to your city, conference, or lab — host a local showdown and put it on the map.\n\nDec 1–3, 2026 · [at Paired Summit](https://pairedsummit.org/)\n\nDate TBA\n\nFAQ\n\nA livestreamed simulator fight club where MicroDuck control policies enter a robot sumo bracket and try to win QuackDown.\n\nNo — MicroDuck hardware isn't released yet. If we can get our hands on prototypes we'll fight them; until then the bouts run in the MicroDuck simulator, so teams compete with policies, not fragile hardware.\n\nMicroDuck is Pollen Robotics' 25 cm open-source biped robot, built for makers to train, retrain, and deploy learned behaviors.\n\nTeams write their own control policies and can use reinforcement learning to train ring control, pushing, dodging, and recovery in simulation.\n\nEight teams enter a single-elimination chart. Everyone starts in the first stage, so no team gets a bye and every policy has to fight its way to the belt.\n\nBest of 3. Stay in the circle, force the other policy out, and keep the controller inside the approved simulator interface.\n\nYes. QuackDown streams live on YouTube (channel announced soon) and is built broadcast-first with match cards, replays, overlays, and sponsor placements.\n\nYes. Sponsors can own the mat center, a ring of apron boards, or the back-wall marquee that renames the whole event.\n\nThere is no cash betting. Friendly side wagers can use Codex or Claude quota, bragging rights, or other non-cash dev-world tokens. DuckFightClub does not operate a betting platform, hold wagers, or settle disputes. Quack responsibly.", "url": "https://wpnews.pro/news/pro-wrestling-for-reinforcement-learning-ducks", "canonical_source": "https://duckfight.club", "published_at": "2026-09-09 23:00:58+00:00", "updated_at": "2026-09-09 23:16:06.940380+00:00", "lang": "en", "topics": ["robotics"], "entities": ["Pollen Robotics", "MicroDuck", "QuackDown", "Golden Beak Belt", "Stanford", "Berkeley", "Paired Summit"], "alternates": {"html": "https://wpnews.pro/news/pro-wrestling-for-reinforcement-learning-ducks", "markdown": "https://wpnews.pro/news/pro-wrestling-for-reinforcement-learning-ducks.md", "text": "https://wpnews.pro/news/pro-wrestling-for-reinforcement-learning-ducks.txt", "jsonld": "https://wpnews.pro/news/pro-wrestling-for-reinforcement-learning-ducks.jsonld"}}