Building a robotics portfolio that gets you hired

The robotics job market in 2026 is overflowing with candidates who took the Modern Robotics Coursera and "know ROS." It's starving for candidates who have built something that moves. The gap between those two groups is almost entirely a portfolio, and portfolios are cheaper to build than a degree. Here's what actually signals.

What hiring managers are scanning for

Three things, in order:

1. Did they finish something? 90% of candidates have half-done projects. The bar is low; clear it.
2. Can they operate end-to-end? Most candidates can only do perception, or only firmware, or only ROS. The rare generalist who can take a project from requirements to a video of the robot actually moving is worth 3× any specialist.
3. Is the engineering honest? Demos that hide failure modes, vague "it uses deep learning" claims, numbers with no denominator — all immediate flags.

Everything below is in service of demonstrating those three.

The hierarchy of portfolio pieces

Tier S: A video of a robot you built doing something useful

30 seconds. Your robot picks up a block, or walks across a desk, or drives around a room. Posted somewhere findable (personal site, Twitter/X, LinkedIn). This single artifact beats any line on a resume. It proves you can execute end-to-end.

Examples worth building: a diff-drive robot that follows you around, a desk-sized robot arm that solves Rubik's, a printed quadruped with a learned gait, a drone that follows a ball.

Tier A: An open-source contribution used by others

You wrote a ROS 2 package. Or contributed non-trivial PRs to MoveIt, Nav2, LeRobot, PyBullet, or Isaac Lab. Or released a library for a specific sensor. GitHub stars help but aren't everything — PRs merged into major projects show you can work in an existing codebase, which is half the job.

Tier B: A writeup that goes viral in the community

A blog post or thread explaining something non-trivial (not restating the docs). "How I got 10× speedup on MPPI," "The three bugs in ROS 2's TF buffer," "I trained a π0 fine-tune on 100 demos and here's what happened." These get shared, and they signal deep work.

Tier C: A high-grade educational credential

Coursera Modern Robotics finished, MIT Underactuated Robotics completed. Nice but below everything above. Every candidate has these now.

Tier D: "Familiar with ROS / Python / PyTorch"

Assumed. Table stakes. Not a signal.

Project ideas that carry weight

Any of these, finished and videoed, will get you interviews:

• Desk-sized quadruped with a learned gait. Cheap to build ($200), hard to do well, demonstrates sim-to-real.
• VLA fine-tune demo. Take OpenVLA or π0, fine-tune on 100 demos of your task, deploy. Shows you understand the modern ML pipeline.
• Full ROS 2 autonomous stack. Small robot, full Nav2 stack, SLAM, autonomous exploration. A classic but still impressive if polished.
• Tactile sensing integration. Build a GelSight or use a commercial one, demonstrate a manipulation task that benefits from it.
• Humanoid teleoperation rig. GELLO clone, ALOHA-style, or your own design. Collect a dataset, show it's clean.
• A custom simulator environment that gets adopted. A new MuJoCo / Isaac Lab task, published on Hugging Face, used by others.

Project anti-patterns

• "I completed the course." The course is the entry ticket, not the destination.
• Tutorial-to-tutorial drift. Fifteen half-finished projects in fifteen different frameworks. Pick one. Finish it.
• A GitHub repo with no README, no video, no demo. If I have to read your code to understand the project, you've lost. Two-minute video + README that explains why you built it.
• "I trained a deep learning model to…" without showing it generalizing. Always report the failure modes.

What "honesty" looks like in a portfolio

• Report success rates with denominators. "93% success over 50 trials" beats "often works."
• Show the failures. A video of the robot dropping the block half the time, with a discussion of why, signals depth.
• Cite the libraries you used. Don't pretend you wrote Nav2 from scratch.
• Scope the claim. "Works on rigid blocks of known geometry" is more useful than "works on anything."

Interviewers have seen every flavor of overhyped project. Calibrated claims are vanishingly rare and extremely memorable.

Where to host

• Personal site. Simple HTML or a free Next.js template. Project per page: what, why, video, details, source link. Hosted on your own domain.
• GitHub. One repo per project, polished README with a GIF at the top. Pinned repos on your profile.
• Twitter / X / Bluesky. Project announcements with videos go far. The robotics community is real and active there.
• YouTube. For longer builds, a 5-minute build video is almost required. James Bruton's channel is the reference.

The conversations that actually get you hired

Hiring managers rarely read whole repos. They read the README, watch the video, and then talk to you about the project in the interview. If you can fluently discuss the design tradeoffs you made, the failures you debugged, and the next steps you'd take — you're in. If you can't answer "what's the biggest weakness?", the polish won't save you.

Build fewer things. Understand them deeply. Have the story ready.

Where the bar actually is

A working desk-sized robot with a clear video, an honest README, and a blog post explaining the design choices is above the 90th percentile of robotics applicants in 2026. Most people never ship. Ship.

A ninety-day plan

1. Weeks 1–3: pick a specific project (narrow scope). Get the simulator version working end-to-end.
2. Weeks 4–8: hardware build. Accept that the first version will be bad.
3. Weeks 9–11: integration. Debug. Measure performance honestly. Iterate.
4. Week 12: record a video. Write a blog post. Polish the README. Post to Twitter/LinkedIn.

Twelve weeks of focused evenings is the price of a portfolio piece that separates you from 90% of applicants. It's cheaper than a degree and worth more in this market.