Loops HouseOpenGrasp

NEAR HACKATHON
Team
1 member- AROwner
Arrijalu
Overview
OpenGrasp is a browser teleoperation game with a real economy. You drive a Franka Panda robot arm (MuJoCo physics, real contact) through expedition tasks on a treasure-map hub: rescue a fragile relic, restore a monolith upright, restock a potion shelf. Every session is recorded at 30 Hz as a robot-training episode in LeRobot format — playing the game literally produces the dataset.Teleoperation games are the cheapest way to collect manipulation data at scale, but incumbents keep their points conversion and licensing economics in a private spreadsheet. OpenGrasp answers with radical transparency: the quality formula, the acceptance threshold (60), the base points and the 70/20/10 licensing split (contributors / treasury / ops) are published constants in POLICY.md, enforced by a Rust policy contract live on NEAR testnet (opengrasp.opengraspdev.testnet). Anyone can re-run the scoring on an episode file and get the same number.It already works end to end. Dataset v0 is public on Hugging Face: 15 episodes, 9,092 frames, LeRobot parquet, MIT — and all 15 are registered on-chain (2,521 quality-weighted points). Each task allows up to 5 verified runs from jittered starting poses, so the data actually varies; practice mode is unlimited and never earns.No token launch or sale in this build: rewards and licensing settle in NEAR, and any future utility token publishes its supply, allocation and vesting first. Season 1 is already charted on the hub map — a second robot arm, two new sites, dataset v1 (the first 100 human-player episodes) and dataset licensing live on mainnet. Play it now: https://jusst01.github.io/opengrasp/v2/