paddy-data-sample / index.html
paddy-data's picture
Upload paddy-data-sample (2026-11-16)
259137d verified
Raw History Blame Contribute Delete
16.2 kB
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Paddy Data Sample</title>
<meta name="description" content="Bimanual cable sorting recorded by teleoperation on xArm7 arms: synchronized cameras, lossless aligned depth, joint states, force-torque and joint commands, with failures labelled, in LeRobot format.">
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500;600&family=Inter:wght@400;500;600;700&family=Space+Grotesk:wght@500;600;700&display=swap" rel="stylesheet">
<link rel="stylesheet" href="style.css">
<link rel="icon" href="data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 32 32'%3E%3Crect width='32' height='32' rx='8' fill='%231d7348'/%3E%3Cpath d='M7 21h18M9 16h14M11 11h10' stroke='%23fff' stroke-width='3' stroke-linecap='round'/%3E%3C/svg%3E">
</head>
<body>
<nav class="nav">
<div class="wrap">
<a class="brand" href="#/">
<svg class="mark" viewBox="0 0 32 32" aria-hidden="true"><rect width="32" height="32" rx="8" fill="var(--accent)"/><path d="M7 21h18M9 16h14M11 11h10" stroke="var(--accent-ink)" stroke-width="3" stroke-linecap="round"/></svg>
Paddy <small>data sample</small>
</a>
<div class="links" id="nav-links"></div>
<button class="theme-btn" id="theme" title="Toggle light/dark" aria-label="Toggle light or dark theme">◐</button>
</div>
</nav>
<div id="view"></div>
<!-- ============ Sections. Each page in app.js (PAGES) picks a list of these. Elements with
data-only="main|details" appear on that page only. Every number comes from data/index.json
or data/schema.json: <span data-f="key"></span> is filled with INDEX.facts[key]. ============ -->
<template id="sec-hero">
<header class="hero">
<div class="wrap hero-grid">
<div>
<div class="eyebrow">Bimanual teleoperation · real hardware</div>
<h1>Robot data that <em>holds up</em> under inspection.</h1>
<p class="lede">One bimanual task recorded untrimmed on two <span data-f="dof"></span>-DoF arms, failures included:
<span data-f="cameras_full"></span> cameras with depth, joints, poses, wrist force-torque and
<span data-f="servo_hz"></span> Hz joint commands on one clock. A sample of every episode, and a full release of a
few with everything as recorded, lossless. <span data-f="format"></span>.</p>
<div class="btns">
<a class="btn primary" href="#/explore">Explore the episodes →</a>
<a class="btn" href="#/#download">Get the data</a>
<a class="btn" href="#/details">Full specification</a>
</div>
</div>
<div class="mosaic" id="mosaic"></div>
</div>
<div class="wrap"><div class="stats" id="stats"></div></div>
</header>
</template>
<template id="sec-spec-head">
<header class="spec-head">
<div class="wrap">
<a class="back" href="#/">← Back to the overview</a>
<div class="eyebrow">Full specification</div>
<h1>The details behind the sample</h1>
<p class="lede">What the overview leaves out: every episode and subtask, the rig and its calibration, every column with
its unit and frame, the per-episode record, the file layout, the conventions, known limits and code to load it.</p>
<nav class="toc" aria-label="On this page">
<a href="#/details#tasks">Subtasks &amp; episodes</a><a href="#/details#rig">Rig &amp; calibration</a>
<a href="#/details#streams">Streams &amp; depth</a><a href="#/details#features">All features</a>
<a href="#/details#episode-metadata">Episode metadata</a><a href="#/details#files">Files</a>
<a href="#/details#conventions">Conventions</a><a href="#/details#quality">Quality &amp; limits</a>
<a href="#/details#code">Load it</a>
</nav>
</div>
</header>
</template>
<template id="sec-overview">
<section id="overview">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">At a glance</div>
<h2>What's in the sample</h2>
</div>
<dl class="glance" id="glance"></dl>
</div>
</section>
</template>
<template id="sec-tasks">
<section class="alt" id="tasks">
<div class="wrap">
<div class="section-head">
<div class="eyebrow" data-only="main">The task</div>
<h2 data-only="main">One bimanual task, labelled to the subtask</h2>
<div class="eyebrow" data-only="details">Subtasks and episodes</div>
<h2 data-only="details">Every step, every episode</h2>
<p><q id="task-instruction"></q> The left arm picks a cable, passes it to the right arm, and the right arm sorts it by
color into the two compartments of the target bin. The operator marks each step live with a foot pedal while
driving; each clip shows one step of one episode in all four views.</p>
</div>
<div class="taskgal three" id="subtask-gallery"></div>
<h3 class="subhead">The scenes</h3>
<div class="scenes" id="scenes"></div>
<div data-only="main">
<h3 class="subhead">Failures stay in, marked at the step that failed</h3>
<p class="fine sub-lede" id="failure-lede"></p>
<div class="taskgal three" id="failure-gallery"></div>
</div>
<div data-only="details">
<h3 class="subhead">Subtasks</h3>
<div class="tbl-wrap"><table class="tbl" id="subtask-table"></table></div>
<h3 class="subhead">All episodes</h3>
<div class="tbl-wrap"><table class="tbl" id="episode-table"></table></div>
<p class="fine">Steps: ✓ success, ✗ failure, in order. Lag: how long after the operator's TCP target the measured
arm follows it; most of it is the arm's own controller, and the hand controller's smoothing is not in it (see
Command lag under “Know before you train”). Live command: share of rows where the arm follows a live teleoperation command
(<code>paddy.action_valid</code>); it is lower while an arm waits for the other arm's turn and while the operator
pauses. This sample was not recorded by a trained expert operator. Click a row to open the episode.</p>
</div>
</div>
</section>
</template>
<template id="sec-overlay">
<section id="model-overlay">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Calibration you can see</div>
<h2>The robot model lands on the robot</h2>
<p>The green shapes are not tracked or fitted to the video. They are the arms, grippers and force-torque sensor of
the URDF that ships with the dataset, posed at each row's measured joint angles and projected through each
camera's calibration in <code>meta/cameras.json</code>, lens distortion included. If the kinematics, a calibration or
the time alignment were off, the outlines would slide off the arms.</p>
</div>
<div class="overlay-grid" id="overlay-clips"></div>
<p class="fine" id="overlay-note"></p>
</div>
</section>
</template>
<template id="sec-rig">
<section id="rig">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">The rig</div>
<h2>One bimanual station, every sensor on one clock</h2>
<p>Two arms mounted <span data-f="mount"></span> on one frame reach into shared bins. A fixed overhead camera sees the
whole scene, a fixed scene camera looks at the bins and the handover from the side, and a camera on each wrist sees
the grasp up close. The operator drives each arm with one VR hand controller.</p>
</div>
<div class="rig">
<div class="rig-views" id="rig-views"></div>
<div class="parts" id="parts"></div>
</div>
<h3 class="subhead" id="calibration">Camera calibration</h3>
<p class="fine" style="margin-top:0">Intrinsics describe the video as delivered (it is not undistorted). Extrinsics place
each camera's color optical frame in an arm's frame and were measured against a ChArUco board; the record of each
calibration ships in <code>meta/cameras.json</code>, and a wrist camera's pose at every row is in
<code>paddy.camera_pose.&lt;camera&gt;</code>.</p>
<div class="tbl-wrap"><table class="tbl" id="calib-table"></table></div>
</div>
</section>
</template>
<template id="sec-streams">
<section class="alt" id="streams">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">What every episode holds</div>
<h2>Observations, actions and the operator's intent</h2>
<p>Each <span data-f="fps"></span> fps row carries the latest reading of every stream at or before its time. Nothing is
resampled into the past and nothing is dropped: where a camera missed a slot, the row says how old its frame is.</p>
</div>
<div class="streams" id="stream-cards"></div>
<div class="depth-demo">
<video id="depth-video" autoplay muted loop playsinline preload="metadata" aria-label="Overhead color video beside its aligned depth"></video>
<div>
<h3>Depth that lines up with the pixels</h3>
<p>Every camera records depth aligned to its color image, so a pixel in the video has a distance in the depth file.
The full release ships it lossless at each color image's size and rate (<span data-f="depth_res"></span> for the
overhead camera, <span data-f="fps"></span> fps) as <span data-f="depth_bits"></span>-bit millimetres, with each
camera's own <span data-f="depth_raw_res"></span> depth image beside it; 0 means no measurement. <span id="depth-fill"></span></p>
<div class="legend-bar"></div>
<div class="legend-ticks" id="depth-ticks"></div>
<p class="fine" id="depth-caption"></p>
</div>
</div>
</div>
</section>
</template>
<template id="sec-features">
<section id="features">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Features</div>
<h2 data-only="details">Every column, with its unit and frame</h2>
<h2 data-only="main">Key features</h2>
<p data-only="details">All features in <code>meta/info.json</code> of the full release, grouped. Units, frames and element
names are read from the dataset itself; open a row's names to see what each element of a vector is. Rows marked
<span class="chip full">full only</span> are not in the three-camera release.</p>
</div>
<div class="tbl-wrap"><table class="tbl feat" id="feature-table"></table></div>
<p class="fine" data-only="main"><a href="#/details#features">All <span id="feat-count"></span> features, the episode metadata and the conventions →</a></p>
</div>
</section>
</template>
<template id="sec-episode-meta">
<section class="alt" id="episode-metadata">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Episode metadata</div>
<h2>One record per episode</h2>
<p>Besides LeRobot's own episode index (<code>meta/episodes/</code>, which also carries the outcome, scene, labels
and episode id), each episode has a sidecar, <code>meta/paddy/episodes/&lt;episode&gt;.json</code>, with its labels,
subtasks, measured rates, lags and stream quality. The fields most worth knowing:</p>
</div>
<div class="tbl-wrap"><table class="tbl" id="episode-fields"></table></div>
</div>
</section>
</template>
<template id="sec-files">
<section id="files">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Files</div>
<h2>Directory layout</h2>
<p>Standard <span data-f="format"></span> at the top level, so <code>LeRobotDataset</code> loads either release as is; everything extra
sits beside it.</p>
</div>
<div class="grid-2 files-grid">
<div class="card"><div class="tree" id="tree-full"></div></div>
<div class="tbl-wrap"><table class="tbl" id="files-table"></table></div>
</div>
</div>
</section>
</template>
<template id="sec-conventions">
<section class="alt" id="conventions">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Conventions</div>
<h2>Time, frames and units, written down</h2>
<p>The rules every column follows, from <code>meta/paddy/info.json</code>.</p>
</div>
<div class="grid-3" id="conv-cards"></div>
</div>
</section>
</template>
<template id="sec-quality">
<section id="quality">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Why you can trust it</div>
<h2>Measured, not assumed</h2>
<p>Every number below is computed from the dataset and recorded in its metadata, per episode.</p>
</div>
<div class="trust" id="trust"></div>
<div class="note" id="stream-note"></div>
<h3 class="subhead">Know before you train</h3>
<ul class="limits" id="limits"></ul>
</div>
</section>
</template>
<template id="sec-code">
<section class="alt" id="code">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Load it</div>
<h2>Three lines to a training batch</h2>
<p>Plain LeRobot for policy training; pandas and PyAV when you want the raw columns and depth.</p>
</div>
<div class="grid-2" id="code-blocks"></div>
</div>
</section>
</template>
<template id="sec-download">
<section id="download">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Download</div>
<h2>Two releases of the same recordings</h2>
<p id="dl-lede"></p>
</div>
<div class="dl" id="dl"></div>
<div class="tbl-wrap dl-compare"><table class="tbl" id="dl-table"></table></div>
<p class="fine" id="dl-note"></p>
</div>
</section>
</template>
<template id="sec-next">
<section class="alt" id="next">
<div class="wrap">
<div class="section-head">
<div class="eyebrow">Coming soon</div>
<h2>This is the first slice</h2>
</div>
<div class="soon-grid">
<div class="soon-card"><span class="chip soon">Coming soon</span><h3>More tasks</h3><p>Cable sorting is the start. New bimanual tasks will join the sample as we record them.</p></div>
<div class="soon-card"><span class="chip soon">Coming soon</span><h3>Hours, not minutes</h3><p>Larger releases in exactly the same format, so code written against this sample keeps working.</p></div>
<div class="soon-card"><span class="chip ok">On request</span><h3>Every episode as recorded</h3><p>The full release has <span data-f="full_episodes"></span> episodes with everything recorded, lossless; the others are available in the same form.</p></div>
<div class="soon-card"><span class="chip ok">On request</span><h3>Your task, recorded</h3><p>Tell us the task, the objects and the scene; we collect it on our rig and deliver it in this format.</p></div>
</div>
</div>
</section>
</template>
<template id="sec-contact">
<section id="contact">
<div class="wrap">
<div class="contact">
<div class="eyebrow">Talk to us</div>
<h2>Need more than a sample?</h2>
<p>We collect robot training data and build the collection rigs that record it. More tasks and
hours-long releases in this same format are on the way, and we record your task on request.
Tell us what your policy needs to learn.</p>
<div class="btns">
<a class="btn primary contact-link" href="#" target="_blank" rel="noopener">Contact Paddy Data →</a>
<a class="btn" href="#/explore">Keep exploring</a>
</div>
</div>
</div>
</section>
</template>
<template id="sec-footer">
<footer><div class="wrap"><span>© Paddy Data · <a class="contact-link" href="#" target="_blank" rel="noopener">Contact</a></span><span id="license-line"></span></div></footer>
</template>
<template id="tpl-explore">
<div class="explore">
<aside class="side">
<div class="filters">
<div class="tabs" id="outcome-tabs"></div>
<input id="search" type="search" placeholder="Filter by scene, outcome or episode" aria-label="Filter episodes">
</div>
<div class="eplist" id="eplist"></div>
</aside>
<div class="main" id="ep-main"></div>
</div>
</template>
<script src="app.js"></script>
</body>
</html>