Aircraft 6-DOF Flight Dynamics Simulator

A flight-dynamics model becomes a useful foundation for autonomy when its equations, assumptions, steady states, disturbances, numerical methods, and verification boundaries remain inspectable.

Most flight-simulation work can hide the underlying model behind a black-box engine. This project builds the nonlinear rigid-body formulation explicitly, then keeps extending the same canonical model with the analysis primitives needed for control: trim, actuator dynamics, reproducible disturbance models, numerical health checks, and local dynamics studies.

• Define aircraft state
• Compute relative air data
• Apply actuator dynamics
• Assemble aero / thrust / gravity
• Integrate nonlinear 6-DOF EOM
• Analyze trim / modes / response
• Export diagnostics

The simulator propagates a 13-state aircraft representation containing local NED position, body-frame velocity, body angular rates, and quaternion attitude. Aerodynamics are assembled from nondimensional stability/control derivatives; relative air velocity includes steady wind and gusts; forces and moments combine aerodynamics, propulsion, and gravity; actuator dynamics sit between commands and achieved deflections; and the equations are integrated with fourth-order Runge-Kutta while preserving quaternion normalization. A reusable trim solver and a dedicated adverse-yaw study now consume the same aircraft derivative instead of copying the physics.

• Public from-scratch 6-DOF FDM with explicit translational and rotational equations of motion.
• Quaternion attitude propagation with normalization and body/NED frame transformations.
• Aerodynamic coefficient build-up includes longitudinal, lateral, directional, rate, and control-surface derivatives.
• Atmosphere, steady wind, one-minus-cosine gust, and reproducible Dryden-style stochastic turbulence models are implemented.
• The reusable trim API supports straight level, climb/descent, coordinated-turn, steady-sideslip, and straight-flight wind conditions; the canonical documentation reports typical convergence in 2–3 iterations near machine precision.
• The adverse-yaw study reuses the canonical trim/simulation path, compares aileron-only and rudder-coordinated cases, sweeps rudder gain, decomposes yaw-moment contributions, and compares nonlinear response with local lateral-directional modal estimates.
• Local study results report Dutch roll at -0.90 ± 3.74j 1/s, roll subsidence at -8.12 1/s, spiral at +0.020 1/s, and a 1.69 s nonlinear yaw-rate period against a 1.68 s linear prediction for the studied case.
• Actuator/state history alignment, quaternion handling, gravity convention, and simulation health guards now have explicit regression/verification coverage.
• The project runner still produces machine-readable CSV/JSON outputs and 16 core engineering plots for the standard simulation case.

• Verification now includes dedicated trim tests, actuator integration/response tests, adverse-yaw study regression coverage, quaternion invariance checks, and simulation health/divergence handling.
• State-aligned actuator reporting prevents aerodynamic reconstruction from using the wrong surface deflection for a propagated state.
• The historical ~2240.6 N·m pitching-moment event remains unresolved because its original state history is unavailable; it is not presented as solved.

• Generic representative aerodynamic coefficients are not a validated aircraft-specific model.
• The adverse-yaw study changes Cn_da to a plausible generic adverse value for mechanism demonstration; its specific numerical predictions are not aircraft-identification results.
• No independent flight-test, wind-tunnel, CFD, or validated reference-simulator correlation has yet established physical predictive accuracy.
• The historical 2240.6 N·m event remains unresolved because its original state history is unavailable.
• Core nonlinear dynamics are reusable and tested, but closed-loop flight-control laws and RL environment integration are still future work.

• Kept the aircraft generic rather than presenting representative coefficients as a validated digital twin.
• Used the canonical aircraft derivative as the shared physics path for simulation, trim, and studies rather than maintaining private copies of the equations.
• Treat actuator rate limiting and lag as physical model behavior, not cosmetic signal smoothing.
• Used quaternion attitude propagation and removed unnecessary q0 ≥ 0 sign forcing because q and -q encode the same physical attitude.
• Added broad numerical health guards as fail-fast sanity checks, explicitly described as generic simulation limits rather than certification boundaries.
• Keep study assumptions explicit: for adverse yaw, Cn_da = -0.03 is a plausible generic value selected to demonstrate the mechanism, not a flight-test or CFD identified coefficient.
• Preserved unresolved anomalies and model-validation gaps as visible limitations instead of forcing a cleaner narrative.

Recover or reproduce the historical anomaly where possible, add independent reference-based validation cases, then build closed-loop control laws and expose the canonical simulator as a reproducible RL environment for aerospace autonomy experiments.

Explore all engineering work →