Aug 2026 → Aug 2028
designed with Claude
No campus, no tuition, no transcript. A two-year, build-first master's in autonomous systems — where the thesis is a crop-scouting drone, the lab is a regenerative microfarm, and every course is whatever the next demo forces me to learn.
Office hours are open. Ask about deadlines, gear, the reading — or anything the syllabus should be able to answer, like "explain SLAM at intuition tier." You can set the clock, too: "today is July 19 2026 — what do I buy?"
The rules that keep a self-issued degree from becoming a reading list.
Pixhawk + ELRS + MAVLink + QGroundControl — the same stack flying research UAVs everywhere. Nearly all of it already on the bench; this year's new-gear budget rounds to zero.
One aircraft, seven demos, thirteen months, all flown over the Treehouse Pantry microfarm. Each milestone names what it ships, what it forces me to learn, who in nature solved it first — and when it's due.
Flight log — first takeoff
August 2, 2026. Day two of the semester, and the X500 is off the bench and off the ground — M0's heartbeat wasn't due for another eight weeks.
A takeoff isn't the exam, though. Kestrel's bar is two minutes of held hover and a landing on purpose — the difference between a leap and flight.
Ship Firmware pinned (the PX4-vs-ArduPilot question settled for good), all sensor calibrations green, ELRS receiver soldered/bound, motors identified and spinning the right directions — props off.
Forces flight-controller architecture · UART & CRSF · PWM/ESC basics · sensor calibration · power budgeting
Nature Metabolism precedes behavior — every organism budgets energy before it budgets movement.
Ship Two minutes of stable manual hover on STOCK arms in an open field, controlled landing, repeatable. FAA registration + TRUST done before the first flight.
Forces flight dynamics (intuition tier) · attitude control loops · LiPo field safety · pre-flight checklists
Nature A kestrel wind-hovers by closing the same feedback loop the autopilot runs — sense, correct, hold.
Ship GPS position hold in a breeze, return-to-launch triggered and proven from altitude, and the custom 3D-printed arms qualified against the stock-arm hover baseline (vibration + hover-current comparison).
Forces GPS & EKF intuition · coordinate frames · failsafes · printed structures: layer orientation, stiffness vs. toughness
Nature Desert ants path-integrate thousands of steps and walk home in a straight line.
Ship One-button autonomous flight over the Treehouse Pantry microfarm: takeoff → waypoints → land, geofenced, mission uploaded over the SiK link, RC only as override.
Forces MAVLink · mission planning · geofencing · telemetry links · flight-time energy math
Nature A honeybee's waggle dance is a waypoint upload — direction and distance, transmitted to another flier.
Ship Camera payload flown on a survey pattern; overlapping photos stitched into a georeferenced orthomosaic of the microfarm — my first map made from the air.
Forces camera geometry · photogrammetry/SfM (SLAM's free cousin) · ground sample distance · image pipelines
Nature Hippocampal place cells — brains invented the map-from-motion trick first.
Ship A repeatable scouting mission flown on schedule; dated map series with plant-health flags landing on the ambient display — the microfarm's history, from above.
Forces GIS & spatial data · change detection · vegetation indices (intuition tier) · telemetry pipelines
Nature Vultures don't search randomly — soaring scouts sweep structured patterns and share what they find.
Ship The full loop with no hands: scheduled mission → flight → map → dashboard. Full writeup — the Year 1 thesis defense, where Year 2's shape gets decided.
Forces mission robustness · failure modes & field ops · program review: what did the farm actually need?
Nature Foraging theory — when to sweep in rows vs. when to wander (Lévy walks).
Undated, guilt-free, and a toy on purpose. The ground track's perception sandbox.
The Bambu Lab CyberBrick forklift (already built and functional), converted into a tunnel-row plant-photography scout.
petsi (a Raspberry Pi 5) is the brain — ArduCam on a swivel servo, WiFi, telemetry to the ambient display. The CyberBrick Core (ESP32-C3, MicroPython) stays on as the actuator controller: 2 DC motor channels + 4 servo ports, commanded over its USB-C serial port. That's the companion-computer pattern — the same architecture as Pixhawk + ground station, at 1/50th the price.
Scope, honestly: Toy-tier drivetrain (plastic 1:48 gears, 030 motors, no encoders) — it will never push through mulch, and that's fine. The drivetrain is disposable; the perception stack, serial protocol, and data pipeline all transfer to Year 2's real rover.
Oct 2027 → Aug 2028. Deliberately undecided until the M6 thesis defense — a syllabus written before its evidence would be guesswork.
The standing bet: the real ground rover— inheriting Rover Zero's perception stack and the kepler452b gas-sensor payload — with the old bench (Jetson, RTK, 3D LiDAR) unlocked item by item behind a purchase gate: a demo must demand it.
The capstone bet stays: an air–ground team. The UAV spots an anomaly from altitude; the rover drives out to investigate. Two robots, one nervous system — soaring scouts and ground foragers have run this division of labor forever.
Dessert track:OpenDuck V2 as a bipedal embodied-AI pet — built only if I'm ahead of schedule, skipped guilt-free if not.
Eight books, two years. Chosen to shape taste, not to transfer syllabus.
Every concept, four questions. The green one is the thesis of the whole school.
“Nature” runs in two registers here, and the program studies both:
Physics, chemistry, electromagnetics, thermodynamics. Newton's third law is as natural as a nervous system. The rover teaches these whether I plan to or not: the motors are an electromagnetics course, the battery is thermodynamics, the LiDAR is wave optics, the RTK fix is relativity in production.
Biology is a billion years of field-tested engineering within those laws — the green rows on every milestone. Ants, place cells, and moth plumes aren't decoration; they're prior art from the longest-running R&D lab in existence.
Progress ledger: every 6–8 weeks, one shipped demo + one journal entry. Fourteen entries over two years is the transcript. There is no other grade.
Two robots in the field, fourteen writeups in the journal, and the ability to hold my own with a lab PI, a farm director, or a robotics CEO — in their language, about their problems.
The original AI-drafted plan was a 5,000-hour syllabus sold as a 1,200-hour one. These didn't make it.
The original v1 syllabus opened with a $1,400 ground-rover shopping list while an X500 drone kit sat ready to fly. Resequenced July 2026 — the rover moved to Year 2 and its budget went to $0. The plan's own build-first rule made the call.
Zero connection to the mission, maximum regulatory drag (Part 107, BVLOS waivers). It was on the list because it sounded cool.
The canon is 8. The rest stay on call — any build that makes me hungry for one earns it back. Just-in-time applies to reading too.
Pillars imply coverage obligations, and coverage guilt is what kills self-study. Kept as a map for naming what a build just taught me — never as a syllabus.
Four "genuine research platforms" at once was fantasy. One serious robot at a time; each inherits the last one's stack. (Rover Zero doesn't count — it's a toy on purpose.)
est. July 2026 · v2, revised July 2026 — resequenced around the hardware in hand · designed in conversation with Claude