Engineering-Workbench / index.html
devkumarhf's picture
Route visitors to the reusable robot policy and measured experiment
eae859d verified
Raw History Blame Contribute Delete
11.1 kB
<!doctype html>
<html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><meta name="description" content="Dev Kumar's engineering workbench: source-linked projects, interactive RST synthesis and original UR7e training-log analysis."><title>Engineering Workbench · Dev Kumar</title><link rel="stylesheet" href="style.css"><script type="module" src="app.js"></script></head>
<body><aside aria-label="Trained robot policy" style="padding:14px 20px;margin:18px auto;max-width:1180px;border:1px solid #496052;border-radius:10px;color:#dbe8df;font-size:14px">New: <a style="color:#a5cfb4" href="https://huggingface.co/devkumarhf/ur7e-sac-line-following" target="_blank" rel="noopener">UR7e trained SAC model</a> · <a style="color:#a5cfb4" href="https://huggingface.co/spaces/devkumarhf/UR7e-Policy-Lab" target="_blank" rel="noopener">Live policy inference</a> · <a style="color:#a5cfb4" href="https://www.kaggle.com/code/devkumarkaggle/ur7e-sac-vs-dls-reproducible-policy-benchmark" target="_blank" rel="noopener">Measured CPU benchmark</a></aside><a class="skip" href="#main">Skip to content</a><header><a class="brand" href="#studies" aria-label="Dev Kumar Engineering Workbench"><svg width="37" height="23" viewBox="0 0 100 55" aria-hidden="true"><defs><linearGradient id="infinity"><stop stop-color="#80d9b3"/><stop offset=".55" stop-color="#a89bdf"/><stop offset="1" stop-color="#d9a07c"/></linearGradient></defs><path d="M50 28C30-7 5 0 5 28s25 35 45 0S95-7 95 28 70 63 50 28" fill="none" stroke="url(#infinity)" stroke-width="7"/></svg><span>DEV KUMAR<span class="brand-sub">ENGINEERING WORKBENCH</span></span></a><nav aria-label="External profiles"><a href="https://dev-kumar.com/" target="_blank" rel="noopener">Portfolio ↗</a><a href="https://github.com/devkumar-projects" target="_blank" rel="noopener">GitHub ↗</a><a href="https://orcid.org/0009-0008-2279-9163" target="_blank" rel="noopener">ORCID ↗</a></nav></header>
<main id="main"><section class="hero"><div><p class="eyebrow"><span class="dot"></span> OPEN ENGINEERING / DEV KUMAR</p><h1>Inspect the work.<br><em>Explore the reasoning.</em></h1><p class="intro">Robotics, perception and control, connected to the code, data and decisions behind each project.</p><p class="available">Available for engineering roles · Île-de-France & Switzerland</p></div><div class="hero-diagram" aria-hidden="true"><svg viewBox="0 0 300 240"><g stroke="#29332f" fill="none"><path d="M0 180H300M0 120H300M0 60H300M60 0V240M120 0V240M180 0V240M240 0V240"/><circle cx="150" cy="120" r="80"/><circle cx="150" cy="120" r="45"/></g><path d="M12 150C70 150 62 20 130 52S183 225 230 139 265 100 291 111" fill="none" stroke="#80d9b3" stroke-width="2"/><path d="M12 157C70 157 62 28 130 60S183 231 230 147 265 108 291 119" fill="none" stroke="#a89bdf" stroke-width="1" stroke-dasharray="3 5"/><circle class="signal" cx="150" cy="120" r="7" fill="#80d9b3"/><text x="14" y="226" fill="#88958c" font-size="10" letter-spacing="2">MODEL → OBSERVE → VERIFY</text></svg></div></section>
<div class="section-nav" role="tablist" aria-label="Workbench views"><button role="tab" id="tab-studies" aria-controls="studies" aria-selected="true" data-tab="studies">01 <span>Project studies</span></button><button role="tab" id="tab-control" aria-controls="control" aria-selected="false" data-tab="control">02 <span>Control Lab</span></button><button role="tab" id="tab-data" aria-controls="data" aria-selected="false" data-tab="data">03 <span>Run explorer</span></button><a class="icp-link" href="https://huggingface.co/spaces/devkumarhf/ICP-Lab" target="_blank" rel="noopener">Open 3D ICP Lab ↗</a></div>
<section id="studies" role="tabpanel" aria-labelledby="tab-studies"><div class="section-heading"><div><p class="eyebrow">FROM PROBLEM TO EVIDENCE</p><h2>Twelve projects. Different kinds of proof.</h2></div><span id="project-count" class="mono" aria-live="polite">12 projects</span></div><div class="filters"><label class="search"><span>Search projects</span><input id="search" type="search" placeholder="Try EKF, MATLAB, LiDAR…"></label><label class="domain-filter"><span>Discipline</span><select id="domain"><option value="all">All disciplines</option><option>Robotics</option><option>Perception</option><option>Control</option><option>Engineering</option><option>Software</option></select></label></div><div class="project-layout"><aside id="project-list" aria-label="Project selection"></aside><article id="project-detail" aria-live="polite"></article></div><p id="empty" hidden>No projects match these filters. Try another method or discipline.</p></section>
<section id="control" role="tabpanel" aria-labelledby="tab-control" hidden><div class="section-heading"><div><p class="eyebrow">A LIVE ALGEBRAIC EXPERIMENT</p><h2>Place the poles. Inspect the controller.</h2></div><span class="pill">Reduced linear model</span></div><p class="section-intro">An independent browser translation of the project's MATLAB synthesis. Every change re-synthesizes the nominal controller; this is not a fixed-controller robustness test.</p><div class="lab-layout"><div class="panel knobs"><label>Pendulum design length L <output id="length-value"></output><input id="length" type="range" min="0.16" max="0.6" step="0.01" value="0.35"></label><label>Dominant pole parameter β <output id="beta-value"></output><input id="beta" type="range" min="0.2" max="3" step="0.1" value="0.5"></label><label>Fast pole parameter γ <output id="gamma-value"></output><input id="gamma" type="range" min="4" max="40" step="1" value="30"></label><label>Probe frequency ω <output id="omega-value"></output><input id="omega" type="range" min="0" max="5" step="0.05" value="1"></label><div class="presets"><button id="initial-design">Initial: β = 1, γ = 10</button><button id="retuned-design">Retuned: β = 0.5, γ = 30</button></div><div class="equation">P(s) = −α² / (s² − α²)<br>α = √(g/L), g = 9.81 m/s²<br>A<sub>cl</sub> = AS + BR<br>H(s) = BT / (AS + BR)</div><a href="https://github.com/devkumar-projects/RST-controller-for-an-inverted-pendulum/blob/0a8a383810060c9c4d79e0ba39a400f11a3acb09/matlab/rst_pendulum_synthesis.m" target="_blank" rel="noopener">Read the source synthesis ↗</a></div><div class="panel"><div class="plot-heading"><h3>Closed-loop pole assignment</h3><span class="mono">s-plane / s⁻¹</span></div><canvas id="poles" width="900" height="360" role="img" aria-label="Complex-plane plot showing open-loop unstable pole and six assigned stable closed-loop poles"></canvas><p class="legend"><span class="green">× assigned closed-loop</span><span class="coral">□ open-loop plant</span><span>Repeated pole at −α (×2)</span></p><div class="stats"><div><small>Bézout relative residual</small><strong id="identity-error"></strong></div><div><small>Gain |H(jω)|</small><strong id="gain"></strong></div><div><small>Phase arg H(jω)</small><strong id="phase"></strong></div></div></div></div><div class="panel frequency"><div class="plot-heading"><h3>Nominal closed-loop frequency response</h3><span>At DC and ω₀ = 1 rad/s, the feedforward design gives H = 1.</span></div><canvas id="bode" width="1200" height="290" role="img" aria-label="Closed-loop magnitude in decibels over angular frequency"></canvas></div><details class="panel"><summary>Inspect R, S, T and the target polynomial</summary><pre id="coefficients"></pre></details><p class="scope">Scope: nominal continuous-time, reduced linear model; fixed internal-model frequency ω₀ = 1 rad/s. No actuator, cart dynamics, saturation, nonlinear stability or hardware validation is simulated. Pole assignment alone is not a robustness certificate.</p></section>
<section id="data" role="tabpanel" aria-labelledby="tab-data" hidden><div class="section-heading"><div><p class="eyebrow">ORIGINAL DATA / REPRODUCIBLE VIEWS</p><h2>Look inside one robot training run.</h2></div><a id="dataset-link" href="https://www.kaggle.com/datasets/devkumarkaggle/ur7e-simulated-robot-line-following-training-log" target="_blank" rel="noopener">Dataset & schema ↗</a></div><p class="section-intro">520 sequential episodes from the public UR7e simulation training log. Explore recorded episode-end summaries, raw values and trailing averages. These are training observations, not an independent evaluation.</p><div class="stats panel"><div><small>Original episodes</small><strong id="episodes">520</strong></div><div><small>Recorded success flags</small><strong id="success-count"></strong></div><div><small>Runs / evaluation splits</small><strong>1 / none</strong></div></div><div class="data-controls"><label>Metric<select id="metric"><option value="r">Episode return</option><option value="progress">Recorded progress (fraction)</option><option value="recent_rmse">Recent tracking RMSE (m)</option><option value="distance">Episode-end distance (m)</option><option value="l">Episode length (steps)</option><option value="is_success">Recorded success flag (0 / 1)</option></select></label><label>Trailing window <output id="window-value">20 episodes</output><input type="range" id="window" min="1" max="80" value="20"></label><button id="download-csv">Download original CSV ↓</button><button id="download-provenance">Download provenance ↓</button></div><div class="panel"><div class="plot-heading"><h3 id="metric-title">Episode return</h3><span>raw observations + trailing mean</span></div><canvas id="training-chart" width="1200" height="400" role="img" aria-label="Original UR7e simulated training episodes with a trailing average"></canvas><p id="hover-readout" class="mono" aria-live="polite">Move across the plot to inspect an episode.</p><div class="legend"><span class="muted">· raw episode</span><span class="green">— trailing mean</span></div></div><details class="panel"><summary>Data provenance and interpretation</summary><p id="provenance"></p><p>The file is unchanged from its pinned GitHub source and keeps the Stable-Baselines3 comment header. <code>recent_rmse</code> is a recent-window tracking summary, not the RMSE of an entire robot evaluation; <code>distance</code> is an episode-end quantity. Sequential observations can be dependent. Changes in rewards or training conditions can affect comparisons.</p><p>There are no random seeds, episode trajectories, observations/actions or held-out test records in this CSV. No confidence interval, causal conclusion or sim-to-real claim is inferred from it.</p><a id="raw-source" target="_blank" rel="noopener">Read pinned original source ↗</a><p>Source licence: MIT; retained in UPSTREAM-UR7E-LICENSE.</p></details></section>
<p id="load-error" role="alert" hidden>The workbench could not load its local data. Please reload the page.</p></main><footer><div>Dev Kumar <span>· Robotics, perception & control</span></div><a href="https://www.kaggle.com/devkumarkaggle" target="_blank" rel="noopener">Executable notebooks ↗</a><a href="https://dev-kumar.com/" target="_blank" rel="noopener">Contact via portfolio ↗</a><span>Browser CPU · no inference API · no uploads</span></footer></body></html>