EDITION 01
COMPLETED
Sep 19–20
Cologne, Germany
The inaugural weekend. Results land on the season board.
SEE MEDIA REPORT →
Teach a robot
to make pizza
in 48 hours.
Choose PizzaBench on standardized hardware or Freestyle with your own robot, model and dataset. Same three challenges — separate leaderboards.
8–12 TEAMSREAL ROBOTSREAL DATAREAL EVALUATION
POLICY pizza_policy_v2
PLACE conf 0.96 · x 310 y 215
PLACE conf 0.96 · x 310 y 215
100
POINTS ACROSS 3 CHALLENGES
2 TRACKS
PIZZABENCH · FREESTYLE
LeRobot
SMOLVLA · ACT · DIFFUSION
0
HARD-CODED TRAJECTORIES ALLOWED
/ UPCOMING EVENTS
Next editions.
Next editions.
Pick your weekend.
PizzaBench teams get identical stations. Freestyle teams bring their own stack. Both run the same challenges — each track has its own season leaderboard.
EDITION 02
UPCOMING
Oct 23–25
Cologne, Germany
Full 48-hour format with the Mystery Pizza championship round. Applications open now.
APPLY FOR THIS EDITION →
EDITION 03
UPCOMING
November
Europe · city TBA
Season finale. Dates and venue announced once the host partner is confirmed.
GET NOTIFIED →
/ THE PREMISE
Most AI hackathons end with a demo on a screen.
This one ends with a robot making pizza. You get a low-cost robotic arm and 48 hours to teach it how to manipulate ingredients — no hard-coded trajectories, no simulated leaderboard. Your model runs on the physical robot.
Pizza is the theme.
Physical AI is the challenge.
REC / PROTOTYPE STATION 01● LIVE FOOTAGE
/ THE PIPELINE
Five steps from
Five steps from
human to autonomy
You will touch the complete embodied-AI stack: robot → teleoperation → dataset → policy → inference → evaluation.
01
Demonstrate
Teleoperate the SO-101 with a leader arm and show it how the task is done.
02
Collect
Turn episodes into a real LeRobot dataset — frames, states, actions, task text.
03
Train
Fine-tune SmolVLA, ACT or a diffusion policy on GPU compute we provide.
04
Deploy
Run inference on the physical arm. It either picks the pepperoni or it doesn't.
05
Compete
Standardized evaluation, then a hidden generalization challenge you cannot train for.
/ ROBOT PIZZA CUP
Two tracks.
Two tracks.
Same challenges.
PizzaBench is the main cup on standardized hardware. Freestyle is for teams and startups with their own robots, models and datasets. Both are judged on the same three challenges — with separate leaderboards.
PIZZABENCH · MAIN CUP
Standardized stack
Assigned SO-101 or ElRobot station, fixed camera, starter repo. Scores count toward the season PizzaBench leaderboard and PizzaBench Champion.
CUSTOM GRIPPERS WELCOME
FREESTYLE
Bring your own
Your robot, your policy, your data. Same three challenges and scoring rubric — ranked on a separate Freestyle leaderboard.
CUSTOM GRIPPERS WELCOME
/ PIZZABENCH CHALLENGES
100 points · scored in both tracks
Precision Topping
30 PTSPick individual pepperoni from the bin and place them accurately on marked target areas. Six pieces, one time limit.
GRASPING · LOCALIZATION
PLACEMENT PRECISION
PLACEMENT PRECISION
Recipe Assembly
30 PTS"Make a pizza with four pepperoni and three mushrooms." Identify, count, sequence — and stop when the recipe is complete.
LANGUAGE-CONDITIONED
LONG-HORIZON EXECUTION
LONG-HORIZON EXECUTION
Mystery Pizza
40 PTSRevealed after models are frozen. New layout, new recipe, shifted bins. No new demonstrations. This one decides each track's leaderboard.
ROBUSTNESS · GENERALIZATION
CHAMPIONSHIP ROUND
CHAMPIONSHIP ROUND
/ THE FINAL
New layout.
New layout.
New recipe.
No new data.
Sunday 15:00: the Mystery Pizza configuration is revealed. Models are already frozen. One policy, one shot, an environment your robot has never seen in exactly this arrangement.
Can your policy generalize?
EVAL_CONFIG.DIFFHIDDEN
- train: 4 pepperoni + 2 mushrooms
+ final: 5 pepperoni + 3 mushrooms
- bin_positions: nominal
+ bin_positions: shifted
- pizza_pose: fixed
+ pizza_pose: sampled(region)
! new_demonstrations: denied
/ WHAT YOU GET
You bring a laptop.
You bring a laptop.
We bring the stack.
No robotics experience required. Every team gets a known-working reference pipeline, so there is always a path to a functioning autonomous policy.
HARDWAREPizzaBench kitSO-101 or ElRobot per team. Freestyle: bring your own.
VISIONFixed RGB camera640×480 @ 30 FPS, identical rig.
COMPUTEGPU accessCloud credits for fine-tuning.
CODEStarter repoBaseline policy + example dataset.
WORKSPACECompetition rigReusable pizza and topping props.
PEOPLEMentorsOffice hours all weekend.
EVALScoring softwareSame protocol for every run.
FUELFood & pizzaEdible ones, made by humans.
YOU BRINGLaptop + curiosityGitHub and Hugging Face accounts.
/ 48 HOURS
Teach. Train. Compete.
Friday you meet the robot. Saturday you fail and improve. Sunday the models are frozen.
FridayBUILD THE PIPELINE
16:00Registration & hardware assignment
17:00Opening session
17:30Robot & LeRobot workshop
18:30Teleoperation & dataset collection
21:00First dataset milestone
SaturdayTRAIN THE ROBOT
09:00Technical office hours
10:00Training & fine-tuning
13:00First autonomous runs
15:00Challenge 1 practice evaluation
21:00Leaderboard checkpoint
SundayCOMPETITION
11:00Models frozen
12:00Official evaluation, C1 & C2
15:00Mystery Pizza revealed
16:00Championship evaluation
17:30Awards, then pizza party
/ AWARDS
Five ways to win
PizzaBench ChampionHighest score on the main-cup leaderboard.
Freestyle ChampionHighest score on the Freestyle leaderboard.
Best VLABest language-conditioned policy.
Best DatasetCleanest demonstrations and docs.
Technical InnovationBest idea in the pipeline.
/ WHO SHOULD JOIN
Teams of 2–4.
Teams of 2–4.
Robotics degree optional.
ML engineers, robotics engineers, computer-vision people, researchers, students, founders and builders. Python and a bit of PyTorch go a long way; everything robot-specific is taught on Friday night.
PYTHON
PYTORCH
COMPUTER VISION
IMITATION LEARNING
FIRST-TIME ROBOTICISTS
/ ORGANIZERS
Bring the robot pizza to your city.
Run your own Robot Pizza Hackathon. Gather your club, community or lab for a weekend, put a robot arm on the table and teach it to make pizza. You bring the room and the crowd. We bring the challenge, the rules and a LeRobot starter repo, so every city plays the same game.
YOU BRING
A space, a crowd of curious builders, and some robot gear if you have it.
WE BRING
The three Robot Pizza Cup challenges, scoring rules, starter repo and practice simulation.
Not sure yet? Tell us what you have. "Nothing yet" is a fine answer.
HOST A CITY →
Takes 2 minutes. No robot experience needed.
CITY HOSTS · 2026 SEASON
COMPLETED
Cologne
Edition 01
SEP 19–20
SEE MEDIA REPORT →
UPCOMING
Cologne
Edition 02
OCT 23–25
APPLY →
UPCOMING
Europe
Edition 03 · city TBA
NOVEMBER 2026
/ PARTNERS
Do I need robotics experience?+
No. We provide the robot stack, a Friday-night workshop, tutorials and a reference implementation that already works end to end.
Do I need my own robot?+
No. Every team is assigned an identical SO-101 or ElRobot station with teleoperation hardware and a fixed camera.
Can I use my own model?+
Yes, provided it satisfies the competition interface and safety requirements. SmolVLA, ACT and diffusion policies are all supported paths.
Can I hard-code the task?+
You may build supporting infrastructure, but the scored autonomous behaviour must satisfy the published competition policy. The Mystery Pizza round is designed to expose memorized trajectories.
Is it real pizza?+
Scored benchmarking uses standardized reusable rigid props so runs are comparable. The live showcase after benchmarking can use edible ingredients.
Why no sauce or dough? Deformable-material and fluid manipulation would sharply increase failure risk for an inaugural event. Future editions can add them.
What happens to the data we collect?+
You keep your code. Dataset licensing and contribution terms are published before registration opens, and public contribution is opt-in.
YOUR NEXT MODEL DOESN'T ANSWER QUESTIONS. IT MOVES THINGS.
Join the Robotics Pizza Hackathon Community
Edition updates, partner news, and a home for robotics and AI builders.