The RobotForge curriculum is complete: 136 lessons, 13 tracks, free forever

When this site launched in April 2026, it had one published tutorial. Today, six weeks later, it holds 136 lessons across 13 tracks — every topic in canonical robotics, every modern AI development, every platform from drones to humanoids, every tool from KiCad to Isaac Lab. The curriculum that took the field 50 years to develop is now end-to-end published in one place. Free. No paywall. Forever.

What's in it

Thirteen tracks, all 100% complete. Each lesson is a 1500–2000-word standalone piece — punchy, information-dense, honest about what works in 2026 and what's still research. Many include hand-drawn SVG diagrams; the math-heavy ones use KaTeX-rendered equations.

• Foundations — the math, code, OS, and tooling every roboticist uses. 10 lessons.
• ROS 2 — the de-facto middleware. Nodes, topics, services, launch, TF, URDF, Nav2, MoveIt. 12 lessons.
• Kinematics & Dynamics — forward, inverse, Jacobian, Lagrangian. The mechanical spine. 10 lessons.
• Control — PID, LQR, MPC, feedback linearization, impedance, force control, Lyapunov, underactuated, real-time. 10 lessons.
• Perception & CV — cameras, LiDAR, YOLO, SAM, VIO, deep CV, VLMs. 10 lessons.
• State Estimation & SLAM — Bayes filter through ORB-SLAM3 and Gaussian splatting. 10 lessons.
• Motion Planning — A* through trajectory optimization, TAMP, behavior trees. 10 lessons.
• Manipulation — grasping, MoveIt, impedance for assembly, dexterous, mobile, tactile. 10 lessons.
• Mobile & Legged — diff-drive through humanoid whole-body control, RL gaits, AV stack. 10 lessons.
• Learning for Robotics — IL, RL, VLAs (π0, OpenVLA, Gemini Robotics), sim-to-real. 12 lessons.
• Simulators — Gazebo, MuJoCo, Isaac Lab, Drake, PyBullet, Webots, URDF/MJCF/USD. 10 lessons.
• Embedded & Hardware — MCUs, encoders, IMUs, power, wiring, Jetson, TinyML, RTOS, PCB, thermal. 12 lessons.
• Frontiers — humanoid whole-body, teleop rigs, tactile, VLA fine-tuning, embodied LLMs, safety, BOM, swarms, soft robotics, careers. 10 lessons.

Estimated total: ~32 hours of reading; far more if you do the exercises. Each track is self-contained but cross-linked.

Why we built it this way

The robotics-education landscape in 2026 has three problems:

1. Fragmentation. Modern Robotics covers kinematics. UPenn does drones. Thrun's Probabilistic Robotics covers SLAM. Each is excellent in isolation; nobody knits them together.
2. Lag. Course curricula update on 5-year cycles. The 2024–26 wave of VLAs, learned gaits, Gaussian-splat SLAM — barely covered in any university course.
3. Cost. The full Coursera + edX + textbook stack runs $2000–$5000. Free-only education means YouTube + arXiv + a lot of context-switching.

RobotForge addresses all three: union of the canonical curriculum, modern AI integrated throughout, free permanently.

What makes it different

Three things, in priority order:

1. The Frontiers track

Topics no other curriculum covers:

• Humanoid whole-body control + retargeting (the Optimus / NEO / Atlas stack).
• Teleop rigs — ALOHA, GELLO, phone-teleop, Vision Pro.
• Production VLA fine-tuning playbook — collect 200 demos, LoRA-fine-tune OpenVLA, deploy.
• Embodied LLM agents — SayCan, Code-as-Policies, the modern stacks.
• Safety + ISO 10218 / 13482 certification.
• Cost-aware BOM engineering.
• Robotics portfolio building — the projects that actually get you hired.

If you're trying to break into 2026 robotics, these are the topics that separate "I took a course" from "I built something that works." No university course covers them yet.

2. The CAD module with MCP

RobotForge has an in-browser CAD (/cad) that ships with a built-in MCP server — meaning your AI agent (Claude Desktop, Cursor, custom) can drive your CAD design directly. Build mechanical parts via dialogue. Export to URDF straight to ROS / Gazebo / Isaac Lab. Watch the design in real-time as the AI mutates it.

No other browser CAD ships this. Built on Manifold (CSG), Three.js (rendering), and the Model Context Protocol (AI tool-calling).

3. Honest tone

Every lesson distinguishes:

• What works in production today.
• What's research-grade in 2026.
• What's still open.

No "this paper just did X so the field is solved" hype. No textbook-isms preserved past their useful life. The tone of a senior engineer explaining the field to a curious junior.

How to use it

Three patterns work, depending on where you start:

If you're new to robotics

1. Start with Foundations. Math, Python, Linux, Git, Docker, frames, units. Skip the math you already know.
2. Pick a robot type (arm, mobile, drone, humanoid) and follow the corresponding sequence: Kinematics → Control → relevant platform track.
3. Add Perception + SLAM when you need autonomy.
4. Add Learning when you want modern AI; add Frontiers for the bleeding edge.

If you're an experienced roboticist hitting modern AI

Read Learning for Robotics end to end (12 lessons; 4 hours). Pair with Production VLA fine-tuning from Frontiers. You'll have working knowledge of the 2024–26 wave by the end.

If you're hiring or evaluating talent

The track structure is also a checklist for what robotics expertise should look like. Each track is a working knowledge domain; full coverage of any track equals competency in that subfield.

What's already shipping besides the tutorials

• /cad — browser CAD with 13-tool MCP server, URDF export, agent-observable viewport, BYO-AI-key support.
• /simulator — in-browser robot physics with Rapier 3D + Three.js. No installs.
• /articles — opinion pieces, deep dives, news.
• /projects — community-shared builds with BOMs.

What's coming next

• CAD Phase B: parametric feature tree with undo/redo, 2D sketcher with constraint solver, OCS2 STEP-import worker.
• Interactive demos: WebAssembly widgets embedded in lessons. Drag a 2-DOF arm; tune a Kalman filter; watch SLAM build a map. Live in the browser.
• Tactile sensing in /cad: simulate grippers with GelSight-style fingertip imprints.
• Comments + community: discuss each lesson; submit corrections; tutorial completion tracking.
• Hardware kits: companion BOMs and paid hardware for the most-built tutorials.

The curriculum is the foundation. From here, the platform grows.

The asks

Four things that would help:

1. Read a lesson. Tell me where I'm wrong. Errata in the comments or a PR link.
2. Share with one person who's learning robotics. Word of mouth is the only marketing this site has.
3. Open an issue for tracks that should grow deeper.
4. If you build something using a lesson, post it under /projects. The platform's value compounds with shared work.

Thanks

To everyone who DM'd "more please" while the curriculum was being filled in. To the open-source robotics community whose work fills these pages. To the people who will use this curriculum to build the next generation of robots — that's why this exists.

We're done with the curriculum. The platform is just getting started.

— RobotForge