On September 2, Ronin (@DeRonin_) published an X article titled "How to become a Robotics Engineer in 6 months (RESOURCES)". It runs to nearly 12,000 words with over 100 linked resources and 27 practical projects. Every price and every link was checked against the vendor or the official docs in September 2026, he says. Where the widely-quoted numbers did not survive checking, he says so instead of repeating them.
Robotics is the least crowded high-value skill in tech right now, he writes. The problem is that almost nobody knows where to start, because the field looks like five fields stacked on top of each other. Some people start with a university robotics textbook and quit at the chapter on rotation matrices, the maths for describing how a part is turned in space. Some buy an Arduino kit, blink an LED, and never build the second thing. Some jump straight into ROS 2 tutorials without knowing what a PWM signal is, a pulse train whose width sets a motor's speed. Others learn only machine learning, which means they can train a policy but cannot make a motor turn. The usual result: months of scattered effort and no working robot to show anyone.
Mid-article, under a heading that reads "Important!!!", he plugs his own YouTube channel. He says the most important skill he realized while writing the article is being able to explain hard technical things in simple words, so he created a channel where he will record videos about AI and robotics engineering. The article embeds a screenshot of the channel with a red circle drawn around the Subscribe button.
He argues for robotics over AI engineering on structural grounds. The models already work, what is missing is someone who can put them in a body. You cannot scrape the physical world; someone has to move a real machine to create the data. There is no Stack Overflow answer for why your gripper keeps slipping, and a robot falling over is not a problem you can hand to a subagent. Nobody clones your robot over a weekend.
He read live job listings from Figure and Skild, and the same four requirements appear in every one: C++ and Python, both, not one or the other; experience on real hardware, since simulation-only does not qualify you as senior; depth in one specialism and literacy across the rest; debugging named as a skill in its own right. Notice what is not on that list, he writes: a specific degree.
Month 1 is electronics, the bench, and the tools you build everything with. Ohm's law, voltage dividers, capacitors, transistors, and reading a schematic, learned in a simulator before spending a dollar. Then soldering, and Python, terminal and Git, because all three are used every week from there to month six. He gives four budget tiers: $0, roughly $45 to $60, roughly $110 to $160, roughly $200 to $300. Verified prices include the Elegoo UNO R3 Super Starter Kit at $42.99, an Adafruit multimeter at $17.50, and a Pinecil V2 iron at $25.99 community, $35.99 retail. His ordering advice: pay the premium for the first kit from Amazon or Elegoo direct so you are building within days, then switch to AliExpress once you know what a 10k resistor is for.
The second month is microcontrollers, motors and sensors. Start on Arduino because every tutorial in existence targets it, then move to ESP32 within weeks. Four motor types matter: brushed DC gearmotor, hobby servo, smart serial bus servo, stepper. One thing to memorize, he says: the L298N driver is in every tutorial and should not be used. It is an obsolete bipolar-transistor H-bridge that drops about 2 volts across its output stage and gets hot. Learn it because the tutorials use it, then switch to the TB6612FNG or DRV8833. For a balancing robot, write a complementary filter before you write a Kalman filter. It is four lines with a constant around 0.98, and it works. The month ends with two builds: a line follower at roughly $38 budget or $105 quality, and a self-balancing robot at roughly $62 or $134.
The third month is mechanical design. Pick one CAD tool and go deep: Onshape Free, with the catch that documents are public on the free tier; Fusion Personal, free on a renewable three-year term; FreeCAD, whose 1.1 release he says is genuinely usable now; or SOLIDWORKS for Makers at $48 a year. Then 3D printing, with verified prices: Ender-3 V3 SE at $199, Bambu Lab A1 mini at $219.99, A1 at $299.99, K1C at $369, P1S at $799. The capstone is the SO-101, an open-source five-joint arm plus gripper from TheRobotStudio and Hugging Face, built as a leader and follower pair so you can hand-guide one and have the other mirror it. Official bill of materials: $229.88 US for the pair, $121.94 for a single follower, excluding printing. That teleoperation setup is what lets you record demonstrations, which is what month six is built on.
Month four is ROS 2, simulation, and building robots the way companies do. ROS 2 is how the industry builds them, and it is the single most requested skill in robotics job listings. ROS 1 is dead. Noetic reached end of life on May 31, 2025, and enormous amounts of highly-ranked tutorial content is still ROS 1. Start on Jazzy Jalisco because almost every course targets it, even though Lyrical Luth is the current LTS. Gazebo Classic reached end of life in January 2025. MuJoCo runs on a normal CPU with no GPU. Isaac Sim's minimum is an RTX 4080 with 16GB of VRAM, and data-centre cards without RT cores such as the A100 and H100 are not supported at all. The practice tasks: your own robot described in xacro, the XML format for describing a robot's parts and joints, simulated in Gazebo with a lidar plugin, driven through ros2_control, mapping a room with SLAM Toolbox and navigating with Nav2.
Month five is the maths that makes robots actually work. PID control, starting with Brian Douglas's free talks. State space, LQR and MPC, including Tedrake's Underactuated Robotics from MIT. Kinematics from the free Modern Robotics preprint. Camera calibration from the official OpenCV docs. Pick and place with MoveIt 2, and understanding planner failure by placing an obstacle in the only viable path.
The last month is robot learning and hireability. LeRobot from Hugging Face, with its teleoperate, record, train, deploy loop. ACT as the starting policy, and Diffusion Policy, which reported a 46.9 percent average improvement over prior methods across twelve tasks. Vision-language-action models, the frontier of robots that take language instructions, with the open-weight ones named: π₀ from Physical Intelligence, pretrained on 10,000-plus hours of robot data; OpenVLA at 7 billion parameters trained on 970,000 episodes from Open X-Embodiment; SmolVLA for affordable hardware. RT-2 from Google DeepMind is historically important and has no public weights. Reinforcement learning via MuJoCo Playground, Isaac Lab, and Levine's CS 285. Then three directions to pick one of: robot learning and embodied AI, autonomy and mobile robotics, or embedded, mechatronics and integration.
The conclusion is where he says the honest version is the only version worth reading. This roadmap will not make you a senior robotics engineer in six months. Senior in this field means five or more years of deploying real-time systems on physical robots, and every job listing he checked says so explicitly. The capital is real: physical AI startups raised $47.4 billion in H1 2026 across 521 deals per Crunchbase, more than 2022 to 2024 combined. Neura Robotics raised $1.4 billion, Skild AI $1.4 billion, Apptronik $520 million, and Figure carries a $39 billion post-money valuation per The Robot Report's 2026 outlook. But hiring lags the capital. North American robot orders grew only 2.0 percent in units in H1 2026 per A3, and US installations fell 9 percent in 2024 per the IFR. The BLS projects 1 to 2 percent growth for the occupational code containing robotics engineers, not the 10 percent figure quoted in career articles with no source behind it.
Pay, with sources given rather than one flattering number: the O*NET/BLS median is $122,930, the only official figure. Entry level roughly $80k to $100k per Payscale and Salary.com. Senior, $160k to $230k at mainstream employers. A live Figure listing for Helix AI Engineer, Robot Learning posts $200,000 to $400,000 base. Teleoperation and data collection averages $28.24 an hour, a genuine foot in the door at frontier companies, but mostly in the most expensive metros in the US, often on-site and physically demanding, and not a comfortable living in San Francisco.
His closing advice: build the projects, do not read about them. Pick one build from each month and actually make it. Write down what broke. Anyone can post a working demo, he writes. Almost nobody documents the four things that failed first and how they diagnosed each one. That write-up is the closest thing to proof that you can actually engineer. Start before you feel ready.