💡 Press Cmd+P (Mac) or Ctrl+P (Windows) to save as PDF. This banner hides when printed.

MissionRobo · Career roadmap

Autonomy / GNC Engineer

Anduril · GA-ASI · Skydio · AeroVironment · Joby

The 16-week path from controls foundations to production GNC (Guidance, Navigation, Control) stacks on autonomous platforms. Curated from public job postings at the top autonomous-systems hiring companies + reading lists shared by senior autonomy engineers.

Intermediate · ~16 weeks · 14 topics · 26 resources

01. Foundations

The math you need before any of the autonomy-specific topics make sense.

Linear algebra for roboticsRequired

Vector spaces, matrix decompositions, SO(3) and SE(3) for rigid-body motion.

Why it matters Every estimator, every controller, every transform between sensor frames is matrix math underneath. Eigen is the C++ library you'll see in every production autonomy stack. What proficiency looks like Junior: comfortable with Eigen API, understand quaternions vs. rotation matrices, can derive the rotation between two coordinate frames. Senior: comfortable in the manifold (Lie algebra of SO(3)/SE(3)) — can implement custom factors in GTSAM, debug numerical conditioning, and explain when to use quaternions vs. axis-angle vs. rotation matrices for a given problem.

3Blue1Brown — Essence of Linear AlgebraFREE — Visual intuition for matrix operations. Best free intro — start here.
Modern Robotics — Lynch & Park — The standard textbook for screw theory and SE(3). Free draft PDFs exist; print copy is worth owning.
Robotics: Kinematics and Mathematical Foundations — Penn / Coursera — paid track with graded problems. Best paid option for SO(3)/SE(3) drills.
Eigen library tutorialFREE — Official C++ docs. Daily reference once you're writing production code.

Probability & random variablesRequired

Multivariate Gaussians, conditional probability, Bayes rule. The math behind every filter you'll write.

Why it matters State estimation IS applied probability. If you can't reason about how uncertainty propagates through a nonlinear function, you can't debug a Kalman filter. What proficiency looks like Junior: know how to multiply Gaussian PDFs, understand the Mahalanobis distance. Senior: derive an EKF or UKF from scratch given the system equations; explain why the Information filter is sometimes preferable in distributed systems.

Probabilistic Robotics — Thrun, Burgard, Fox — The book on Bayesian filtering for robotics. Required reading at every autonomy team.
Stanford CS 228 — Probabilistic Graphical ModelsFREE — Free Stanford lecture notes. Heavy on theory but tightens your conditional-probability fluency.

Rigid body dynamicsRecommended

Newton-Euler equations, equations of motion in body vs. world frame, quaternion kinematics.

Why it matters You'll model a quad / fixed-wing / ground vehicle as a 6-DOF rigid body. Get the frames wrong and your simulator drifts, your control loop oscillates, and the bug is invisible in code review.

Spacecraft Dynamics and Control — Coursera (Schaub) — Hands down the cleanest treatment of rotational dynamics. Aerospace-flavored but transfers directly to multirotor / fixed-wing.
MIT 16.07 — Dynamics (OCW)FREE — Free MIT lectures. Cover Newton-Euler from first principles.

C++ for autonomyRequired

Modern C++17/20 idioms, RAII, smart pointers, build systems (CMake), real-time concerns.

Why it matters ROS 2 is C++17. PX4 is C++14. The big-three GNC employers all hire C++ engineers; Python-only candidates rarely clear the bar for senior IC roles.

A Tour of C++ (3rd ed) — Stroustrup — Short, dense. Best on-ramp for engineers who know C++ but haven't kept up with C++17/20.
CppCon talks — Real-time C++FREE — Search "real-time" — Herb Sutter and Bryce Adelstein-Lelbach's talks are essential for embedded autonomy.

02. Sensing & Estimation

Turn noisy IMU / GPS / camera / LiDAR readings into a clean state estimate.

Extended Kalman filterRequired

EKF derivation, predict/update steps, Jacobian linearization, covariance propagation.

Why it matters The EKF is still the workhorse estimator in production autonomy. You will be expected to implement one in an interview, debug a diverging filter in production, and explain why the UKF or particle filter would or wouldn't help.

Kalman and Bayesian Filters in Python (Labbe)FREE — Free Jupyter notebook book. The interactive derivations make EKF intuition stick.
University of Cyprus EKF LecturesFREE — Cyrill Stachniss — clearest derivation on YouTube. Worth pausing to write the equations yourself.

Visual-inertial odometry (VIO)Required

Tightly-coupled IMU + camera fusion. The sensing stack on every consumer drone and most autonomy testbeds.

Why it matters VIO is what lets a Skydio drone fly through trees, what lets a Boston Dynamics Atlas balance, and what 95% of GPS-denied autonomy depends on.

VINS-Fusion (HKUST)FREE — Open-source production-grade VIO. Reading this code IS the curriculum.
OpenVINS — paper + implementationFREE — MSCKF-based VIO from U of Delaware. Excellent companion to VINS-Fusion for comparison.

SLAM (simultaneous localization & mapping)Recommended

Pose graph optimization, factor graphs (GTSAM), loop closure, lidar vs. visual SLAM.

Why it matters Known SLAM stacks (ORB-SLAM3, LIO-SAM, Cartographer) come up in every autonomy interview. You don't need to implement one from scratch but you should be able to explain trade-offs.

ORB-SLAM3 paper + codeFREE — The reference visual SLAM stack. Reading the paper + running the code is half a master's thesis.
GTSAM Factor Graphs tutorialFREE — Georgia Tech Smoothing And Mapping library. Production-grade factor-graph optimizer used by Skydio + others.

GPS-denied navigationOptional

Dead reckoning, terrain-relative navigation, magnetic anomaly, signals-of-opportunity.

Why it matters This is the defense-specific edge over commercial robotics — Anduril, Shield AI, Skydio defense all hire heavily here. Contested-EM environments mean GPS isn't a given.

GPS-Denied Navigation — JHU APL primerFREE — Open-access overview of GPS-denied techniques used in defense. Good starting context before deep papers.

03. Control Theory

Make the platform actually do what the autonomy stack tells it to.

Classical control (PID + frequency domain)Required

Bode plots, root locus, stability margins. Still the foundation under every modern controller.

Why it matters Frequency-domain thinking IS how you debug a controller that's oscillating in flight. Modern control textbooks tend to skip this; it's the most common gap in junior candidates.

Feedback Control of Dynamic Systems — Franklin, Powell, Emami-Naeini — The standard undergraduate controls text. Read chapters 4-7 for Bode/Nyquist intuition.
Brian Douglas — Control Systems LecturesFREE — Free YouTube series. The single best free controls instructor on the internet.

LQR / state-space controlRecommended

Linear quadratic regulator, observability, controllability, the algebraic Riccati equation.

Why it matters Attitude controllers on most multirotor platforms are LQR underneath. Knowing how to retune one without resimming everything is a senior-IC skill.

Underactuated Robotics — Russ Tedrake (MIT)FREE — Free MIT course. The bridge between classical and modern control, with robotics-flavored examples.

Model predictive control (MPC)Recommended

Receding-horizon optimization, constraint handling, real-time QP solvers (OSQP, acados).

Why it matters MPC is the modern answer for trajectory tracking with state + input constraints. Joby, Skydio, and Anduril all use it in production.

Model Predictive Control — Borrelli, Bemporad, MorariFREE — Free course materials from Berkeley. The textbook is the field reference.
acados — fast MPC frameworkFREE — Open-source code-generated MPC. Used in production at Skydio. Reading the examples is the fastest on-ramp.

04. Production GNC stacks

What separates a research engineer from someone shipping autonomy on real hardware.

Read PX4 and ArduPilot sourceRecommended

The two reference open-source autopilots. Both run on real flight hardware in millions of vehicles.

Why it matters Reading production GNC code is the single highest-bandwidth way to learn how the textbook stuff becomes real. Pick one (PX4 if you're aerospace-focused, ArduPilot if you're multirotor/rover-focused) and trace the attitude loop end-to-end.

PX4 Autopilot — repo + docsFREE — Industrial-grade open-source autopilot. Read src/modules/mc_att_control for the attitude loop.
ArduPilot — repo + docsFREE — The other reference autopilot. Multirotor, fixed-wing, rover, sub — all in one codebase.

Hardware-in-the-loop (HIL) simulationRecommended

Connect the real flight controller to a sim. The only way to validate GNC code before flying.

Why it matters No senior GNC engineer ships without HIL. PX4 + Gazebo, JSBSim, or DJI's sim — pick a stack and become fluent in its quirks.

PX4 + Gazebo HIL setup guideFREE — Official PX4 docs. Working HIL setup in a weekend.

Certification & flight-safety basicsOptional

DO-178C for software, DO-254 for hardware, ASTM F38 for sUAS. The compliance layer that gates Series B+ companies.

Why it matters Every company building autonomy for piloted aircraft, large UAS, or defense customers eventually hits this. Knowing the basics — even if you've never written DO-178C-compliant code — separates you in interviews.

DO-178C primer — RTCA — RTCA runs paid courses. SkyRadar and other vendors offer cheaper introductions if you want a survey.