Robotics is the least crowded high-value skill in tech right now. 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
- 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, then wonder why nothing works on real hardware
- Others go the opposite way and learn only machine learning, which means they can train a policy but cannot make a motor turn
The result is usually the same: months of scattered effort and no working robot to show anyone.
If your goal is to become a robotics engineer, you do not need a mechanical engineering degree and you do not need to derive the dynamics of a seven-link arm by hand. You need to be able to make physical things do what you tell them to, reliably, and then prove it.
That means learning how to:
- read a schematic, wire a circuit, and debug it with a multimeter
- program a microcontroller to drive motors and read sensors in real time
- design and manufacture your own mechanical parts
- build robots the way companies build them, in ROS 2, with simulation
- close a control loop and know why it oscillates when it does
- teach a robot a task from demonstrations instead of hand-coding it
- ship a portfolio that survives three follow-up questions in an interview
Follow: youtube.com/@deronin_23 — It would mean a lot ❤
Why robotics and not AI engineering
Everyone in your timeline is an AI engineer now, and that is the whole problem. You cannot build a moat out of what a subagent does at 3am for pennies. Robotics is the opposite, for reasons that are structural rather than fashionable:
- 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
- a robot falling over is not a problem you can hand to a subagent
- nobody clones your robot over a weekend
AI engineering is crowded and robotics is empty.
What a robotics engineer actually does
A lot of people hear "robotics engineer" and imagine someone designing a humanoid from scratch. In reality the job splits into specialisms, and you pick one and stay literate in the rest:
- perception and computer vision
- controls and state estimation
- motion planning and manipulation
- robot learning
- embedded and firmware
- simulation
- systems and integration
- deployment and operations
I read live job listings from Figure and Skild, and the same four requirements appear in every single 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
Electronics, the bench, and the tools you build everything with
Goal: be able to read a schematic, build a circuit that works, and find the fault when it does not.
Almost every roadmap skips this and starts at Arduino, which is why so many people can copy a wiring diagram but cannot fix anything when it breaks. Robotics is the one software field where the bug is sometimes a loose wire, a sagging battery or a motor drawing more current than your regulator can supply.
1. Electronics fundamentals
You need Ohm's law, voltage dividers, what a capacitor does, how a transistor switches, and how to read a schematic. You do not need to design an op-amp from first principles.
Free, in-browser. Animated electron flow and live voltage colouring — fastest way to build intuition.
Virtual breadboard + Arduino + multimeter together, catches wiring mistakes before you own parts.
Complete open-licensed EE textbook in six volumes. Reference when a video hand-waves.
Short, fast, funny videos sorted into beginner/intermediate/advanced.
Best single paper book for someone who has never held a multimeter.
What to focus on
- Ohm's law and voltage dividers until automatic
- Current draw — why motor stalling browns out your MCU
- Reading schematics: resistor, capacitor, diode, transistor, ground, Vcc
- Pull-up/pull-down resistors
- Decoupling capacitors — why every IC needs one
- Battery basics: LiPo cell counts, C ratings, never leave charging unattended
Build a voltage divider in Falstad, calculate output by hand, confirm in sim. Then build a transistor switch that turns an LED on from logic-level input — that exact circuit will later let a 3.3V pin control something needing more current.
2. The bench, tools & honest budget
| Tier | Cost | What you get |
|---|---|---|
| Tier 0 | $0 | Falstad, Tinkercad, All About Circuits. Learn Ohm's law before spending. |
| Tier 1 | ~$45-60 | Starter kit + multimeter, solderless |
| Tier 2 | ~$110-160 | + soldering iron, solder, cutters, strippers, helping hands, perfboard |
| Tier 3 | ~$200-300 | + bench PSU, better meter, desoldering pump, storage, chassis kit |
Best value pick, most beginner courses target it.
Cheapest real entry if money is tight.
Best curriculum — 16 circuits across 5 projects ending in a working robot.
Volts, current to 20A, continuity, resistance, capacitance — everything you need for years.
Real temp-controlled iron, USB-C powered, takes TS100 / Hakko T12 tips.
Where to order
AliExpress cheapest (3-10x less) but 2-6 weeks, no support. Amazon mid-priced and fast — right for first kit. Adafruit/SparkFun cost more but every page links a full tutorial and support is real. Pololu best for motors/drivers. DigiKey/Mouser for exact parts by spec.
Buy kit + multimeter. Measure battery voltage, resistance of 5 random resistors vs colour bands, then use continuity mode to find a deliberately broken wire you make yourself.
3. Soldering
Iron selection, technique, photos of every failure, safety.
Including what to do when you blow the fuse — you will.
Solder header pins onto a cheap breakout board, test every pin with continuity, repeat 3 times. Desolder one and resolder it — removing components badly is how most beginners destroy boards.
4. Python, terminal & Git
Rigorous, problem sets + final project, structure makes people finish.
Gentlest starting point if CS50P feels steep.
Shell, scripting, command-line fluency robotics runs on.
Best visual tool for branches/merges.
• Read a schematic and build it on breadboard
• Calculate if resistor value is right before plugging
• Find short/break/dead component with multimeter
• Solder clean through-hole joint and check it
• Write Python, run from terminal, push to GitHub
• Explain why motor stalling can reset your MCU
Microcontrollers, motors and sensors, and your first moving robot
Goal: Build a robot that moves, senses, and corrects itself. Line-following robot is a closed control loop with sensor input, actuator output and tuning problem — exactly what a humanoid is, only smaller.
1. Arduino
Start on Arduino rather than ESP32 — ecosystem is enormous and every tutorial targets it. You will move to ESP32 within weeks, nothing wasted.
100+ lessons, homework each — best fit if you failed at Arduino before.
Runnable sketches inside IDE — fastest to “something moved”.
Authoritative reference for digital/analog IO, PWM, I2C, SPI, UART.
6,000+ projects with wiring and code.
Reaction-timer game: LED fires after random delay, button stops clock, time prints to serial. Uses interrupts, debouncing, non-blocking timing with millis() — demonstrable in 15s video.
2. ESP32
Where you go the moment you want WiFi, Bluetooth, more power or two cores. Buy ESP32-S3 as main, plus one classic ESP32 so older tutorial code runs unmodified.
Highest-signal free library, 250+ project index.
Only source of truth once you outgrow Arduino layer.
Bridge that makes Arduino vs ESP-IDF a spectrum.
Long-form, wiring-diagram-heavy, ESP-NOW, OTA, low-power.
| Board | Price | Link |
|---|---|---|
| ESP32-S3-DevKitC-1 8MB | $15.95 | Adafruit |
| Classic ESP32 Dev | $15.00 | Adafruit |
| Seeed XIAO ESP32-C3 | $4.99 | Seeed |
Serve a web page from ESP32 showing live sensor readings + buttons that drive a servo, accessible from phone on same network. Teaches WiFi, HTTP, async work.
3. Motors, drivers & actuation
Four types matter: Brushed DC gearmotor (cheap, needs H-bridge), Hobby servo (180° internal loop), Smart serial bus servo (daisy-chained, position/velocity/current feedback), Stepper (open-loop absolute positioning).
| Driver | Price | Use |
|---|---|---|
| Adafruit DRV8833 | $5.95 | Cheapest good driver |
| SparkFun TB6612FNG | $14.77 | Correct default replacement for L298N |
| Pololu gearmotor w/ encoder | $19.95 | Encoder wiring solved |
| FeeTech STS3215 smart servo | $31.71 | Used in SO-101 arm |
Drive one DC motor forward/backward at 5 speeds via PWM, add encoder, write function that turns wheel exactly one full revolution regardless of battery voltage — your first real closed loop.
4. Sensors & 5. First robots
| Sensor | Price | Notes |
|---|---|---|
| HC-SR04 ultrasonic | $3.95 | Wide cone, poor on soft surfaces |
| VL53L0X ToF laser | $14.95 | Narrow 35° cone |
| MPU-6050 IMU | $12.95 | Cheap classic, you do fusion |
| BNO085 9-DoF | $29.50 | Fusion on-chip, quaternion out |
| RPLIDAR C1 lidar | $69.00 | 360° lidar |
Line-follower: Quality ~$105, Budget ~$38. Balancing robot: Quality ~$134, Budget ~$62. BOM details in original thread — ESP32-S3 + Romi chassis + TB6612FNG + QTR-8RC array.
• Drive motor at controlled speed, know PWM duty vs actual RPM
• Read encoder and close position loop
• Wire I2C sensor from datasheet without tutorial
• Fuse accel + gyro into stable angle
• Explain P/I/D by describing what robot did
• Two working robots on GitHub with wiring + code + failure notes
Mechanical design, CAD and manufacturing your own parts
Goal: Design a part in CAD, manufacture it, and have it fit. This separates people who assemble kits from people who build robots.
CAD — pick one and go deep
- Onshape Free: Browser, works on Chromebook, assembly/mate behaves like robot joints. Catch: docs are public
- Fusion Personal: Free 3-year term non-commercial, CAM + print integration
- FreeCAD 1.1 (Mar 2026): Genuinely usable now, no cloud/payment
- SOLIDWORKS for Makers: $48/year, watermarked files
3D Printing — verified Sep 2026
| Printer | Price |
|---|---|
| Creality Ender-3 V3 SE | $199 |
| Bambu Lab A1 mini | $219.99 |
| Bambu Lab A1 256mm | $299.99 |
| Creality K1C enclosed | $369 |
| Bambu Lab P1S enclosed CoreXY | $799 |
Filaments: PLA/PLA+ prototypes (stiff, creeps, softens 55-60°C), PETG default real part (tough), ABS/ASA hot/outdoor (needs enclosure), Nylon gears, CF filled stiff links (hardened nozzle), TPU feet/bumpers/gripper fingers.
Build a real arm — SO-101
Open-source 5-DOF arm + gripper from TheRobotStudio + Hugging Face, leader + follower pair for teleop and demonstration recording. Official BOM: $229.88 pair, $121.94 single follower excl. printing.
The platform for months 5-6
ROS 2, simulation, and building robots the way companies do
Goal: Build robot in ROS 2, simulate it, make it map a room and navigate autonomously. You can do all of it with no hardware, on normal laptop, no NVIDIA GPU.
Which ROS 2 to install
| Release | Date | EOL | Ubuntu |
|---|---|---|---|
| Lyrical Luth (LTS) | May 2026 | May 2031 | 26.04 |
| Jazzy Jalisco (LTS) | May 2024 | May 2029 | 24.04 |
| Kilted Kaiju | EOL 31 Dec 2026 — do not start here | ||
Canonical reference: nodes, topics, services, actions, tf2, URDF.
Browser-based ROS env with simulated robots — no Ubuntu dual-boot hell. Free tier includes 3 courses.
Best free end-to-end narrative: URDF → sim → ros2_control → Pi + lidar → SLAM → Nav2.
Simulation — naming fixed
Gazebo Classic 1-11 EOL Jan 2025. It was rewritten as Ignition Gazebo, renamed back to Gazebo Apr 2022 (ign → gz). Current: Fortress, Garden, Harmonic, Ionic, Jetty. Pair: Humble+Fortress, Jazzy+Harmonic, Kilted+Ionic, Lyrical+Jetty.
- Gazebo: Learn first, natively wired into ROS 2, runs on laptop
- MuJoCo: Fastest accurate contact dynamics, CPU-first, research standard for locomotion/learning
- Isaac Sim: Photoreal + synthetic data, requires RTX 4080 16GB VRAM min, A100/H100 not supported (no RT cores)
- Skip PyBullet (no release since 2022), Watch Genesis (v1.0 fresh, no ROS 2)
SLAM & Nav2
Launches Nav2 in sim in under 5 min, pre-configured VS Code dev container.
Currently supported ROS 2 SLAM library.
The maths that makes robots actually work
Goal: Understand and implement control and perception underneath everything you've built.
Control theory — PID → LQR → MPC
7 parts: integrator windup, derivative filtering, tuning.
Where control becomes robotics: pendulums, cart-poles, walking, LQR, MPC.
Kinematics & dynamics
Standard textbook, screw theory product-of-exponentials, cleaner than DH.
FK, Jacobian, IK, trajectory, 50+ real robot models.
Perception & manipulation
2-3h from OpenCV themselves.
Every robotics engineer must do this from memory.
Robot learning, specialisation, and becoming hireable
Goal: Pick one direction, build portfolio piece, start applying. You now have whole stack: electronics, embedded, mechanical, ROS 2, control, perception.
Part one: modern robot learning — LeRobot
LeRobot is Hugging Face's PyTorch library for real-world robotics — open hub low-cost robot-learning world converged on. Workflow: teleoperate → record → train → deploy. Drive by hand, save sync video+action, policy learns to imitate, runs on its own.
30-45 min per unit, entirely sim + public datasets.
Focus: record-train-deploy end-to-end, dataset quality (sloppy demos = sloppy policy), ACT (predict chunks), Diffusion Policy (+46.9% avg improvement across 12 tasks).
Record 50 demos single task on SO-101 (pick cube → bin), train ACT, deploy. Will work ~50% time. Record 50 more covering failure cases, retrain. Document success rate before/after — that number is portfolio piece.
Vision-language-action models
| Model | Weights | Notes |
|---|---|---|
| π₀, π₀-FAST, π₀.₅ — Physical Intelligence | Apache 2.0 open, 10k+ hrs data | openpi — transfer not guaranteed, honest warning |
| OpenVLA 7B | Fully open, 970k episodes Open X-Embodiment | Best documented |
| GR00T N1.7 — NVIDIA | Code Apache 2.0, weights NVIDIA Open Model License | Needs 16GB+ VRAM Isaac-GR00T |
| SmolVLA — HF | Compact, for affordable hardware | Right one to fine-tune on SO-101 smolvla_base |
Reinforcement learning
Industry standard legged/humanoid sim-to-real, needs RTX hardware.
Pick a direction
| Direction | Best if | Focus |
|---|---|---|
| 1. Robot learning & embodied AI | Want frontier companies, highest ceiling | LeRobot, VLA fine-tuning, imitation, RL, sim, PyTorch — highest-paid, most competitive |
| 2. Autonomy & mobile | Largest number of jobs | ROS 2, Nav2, SLAM, perception, fusion, C++ — controls + field service = 20%+ postings |
| 3. Embedded/mechatronics/integration | Want to work immediately, contractor | Firmware, motor control, real-time, PLCs, safety — least glamorous, most consistently employable. 66% projects report delays from certification |
Portfolio — what recruiters actually screen
• GitHub repos with logged metrics + commit history showing debugging
• Real-robot deployment with reliability data
• Public datasets on LeRobot Hub
• Contributions to ROS 2 core, Nav2, MoveIt, Isaac Lab, LeRobot
• Systems integration: sensor→actuator→planning→control with video
• Still on EOL ROS 1
• Claimed projects can't survive 3 follow-up questions
• Pure deep-learning, no kinematics/embodiment
• Tutorial completion certificates
Take 3 best projects, rewrite READMEs: video at top, wiring/architecture diagram, actual numbers measured, section “what broke and how I fixed it”. That last section is highest-value — cannot be faked from tutorial.
Interviews
Not software interviews — leetcode weaker predictor. Expect: IK questions, PID/feedback, sensor fusion/Kalman, SLAM concepts, RRT, C++ vs Python tradeoffs. At good companies: simulation debugging (broken MuJoCo/Isaac controller), ROS 2 architecture problem, long conversation about specific past failure.
Glassdoor has 1,721 robotics engineer interview Qs from 877 companies: glassdoor.com
Conclusion — What you can expect after 6 months
This roadmap will not make you a senior robotics engineer in six months. Senior means 5+ years deploying real-time systems on physical robots, every job listing says so explicitly.
What it will make you is someone who can build, debug and ship working robots, who understands stack from electrical layer up to learned policy, and who has physical portfolio to prove it. That is enough to get into the field, and field is where compounding happens.
The demand picture, honestly
The capital is real. Physical AI startups raised $47.4B in H1 2026 across 521 deals, more than 2022-2024 combined per Crunchbase. Narrower robotics-only funding hit $18.8B by late June 2026 vs $15B all of 2025.
Individual rounds: Neura Robotics $1.4B, Skild AI $1.4B, Apptronik $520M, Figure $39B post-money per Robot Report 2026 outlook.
But hiring lagging capital. North American robot orders +2.0% units H1 2026 per A3. US installations -9% 2024 per IFR. BLS occupational code containing robotics engineers projects 1-2% growth, not 10% figure quoted in career articles with no source.
Robotics much smaller labour market than software. BLS: 106,100 annual openings software developers vs ~57,200 for mechanical+electrical+industrial engineers combined.
Accurate claim: money arriving faster than qualified people, shortage concentrated in specific specialisms, entry tier unusually open to people without degrees. Largest robotics job category by posting volume is Automation and Robotics Technician at 20.3% of 3,113 postings, 68% technician roles satisfied by associate's degree or certificate per O*NET.
What it pays
Full-time US — sources disagree, so range + source rather than one flattering number:
| Level | Range | Source |
|---|---|---|
| Median robotics engineer | $122,930 | O*NET/BLS — only official figure |
| Entry | $80k-$100k | Payscale/Salary.com, higher coastal/well-funded |
| Mid 3-6 yrs | $120k-$165k | Conventional employers |
| Senior | $160k-$230k | Mainstream |
| Frontier physical-AI | $200k-$400k base | Figure Helix AI Engineer $200-400k, Perception $200-350k live listing |
| Foundation model specialist | $280k-$475k TC | Recruiter placement data, heavily equity-weighted |
Outside US: Germany ~€70k median total Glassdoor, Switzerland CHF 95k, UK £32k-£50k base, India ~₹653k.
Freelance & contract
- Upwork robotics: $22-$70/hr, clustering $30-$50
- ROS project work: $500-$1,200 node, $1,200-$2,500 launch config, $4,500-$7,000 sensor integration pipeline, $7k-$12k full system architecture
- ZipRecruiter ROS dev avg $52.84/hr
- In robot cell, robot itself ~25% total cost. Other 75% is engineering, safety, integration. $35k arm becomes $80k working system — $45k gap is what you are selling
Entry points needing no degree
- Teleoperation & data collection avg $28.24/hr, most $22.12-$32.93. Figure Humanoid Robot Operator $25-$35/hr no degree, Weave Robotics $25/hr
- Robotics technician median $73,900 per BLS, typically associate's degree
Now what you should actually take away
Build the projects, do not read about them. Pick one build from each month and actually make it. Line follower, balancing robot, arm, ROS 2 navigation stack, trained policy. Break them, fix them, film them, push to GitHub. Every recruiter source checked: portfolio-first sourcing beats keyword search, projects that get people hired have logged metrics and visible debugging history.
Write down what broke. Anyone can post working demo. Almost nobody documents four things that failed first and how they diagnosed each one. That write-up is closest thing to proof you can engineer, reason your repo survives third follow-up question in interview.
Start before you feel ready. Apply for technician role, take teleoperation shift, contribute small fix to Nav2, message founder of three-person robotics startup. Gap between "I am learning robotics" and "I work in robotics" is almost entirely gap of nerve, nobody in field will ever tell you studied enough.
Six months of two focused hours a day is enough to change what you are able to do with your hands and your head. Hardware cheaper than ever, software open, courses free, people who can make machines work genuinely scarce.
Go build something that moves.