Atmospheric Descent Control Analysis

The Challenge of Controlled Descent

SpaceX's Falcon 9 booster executes one of aerospace engineering's most audacious maneuvers: atmospheric entry at Mach 6, followed by controlled descent through 40 kilometers of increasingly dense air, culminating in a precision landing on a drone ship pitching in ocean swells. The margin for error? Effectively zero.

This project simulates that descent trajectory—not to replicate SpaceX's proprietary algorithms, but to explore fundamental control trade-offs across three distinct flight regimes. How do you steer a 40-meter aluminum tube through vacuum, transitional atmosphere, and dense air using different control actuators for each phase?

Three Control Modes, One Descent

The simulation implements altitude-dependent mode switching:

• RCS-Only Mode (>30km): Reaction Control System cold-gas thrusters provide attitude control in near-vacuum. At 40km altitude, atmospheric density is 1/250th of sea level—aerodynamic surfaces generate negligible force.
• Hybrid Mode (10-30km): Transitional regime where both RCS and aerodynamic forces contribute. Control authority gradually shifts from thrusters to grid fins as air density increases exponentially.
• Grid Fins Only (<10km): Below 10km, air is dense enough for pure aerodynamic control. Grid fins—lattice structures that deploy from the rocket's body—provide high drag coefficient with minimal mass penalty.

Physics Modeling

The simulation solves 6-DOF (six degrees of freedom) equations of motion: three translational (x, y, z position) and three rotational (roll, pitch, yaw). At each timestep:

• Atmospheric Density: Exponential model ρ(h) = ρ₀ * exp(-h/H) where H ≈ 8.5km (scale height)
• Aerodynamic Forces: Drag F_d = 0.5 * ρ * v² * C_d * A opposing velocity vector
• Grid Fin Torque: τ = r × F where r is moment arm, F is aerodynamic force on deflected fin
• RCS Torque: Direct torque from thruster pairs: τ = F_thrust * moment_arm

Integration via SciPy's odeint (implicit Adams method for stiff ODEs), timestep 0.01 seconds.

PID Control Architecture

A cascaded PID controller maintains desired orientation. Error signals (difference between target and current attitude) feed into three control terms:

• Proportional: Immediate response proportional to current error
• Integral: Accumulates error over time, eliminating steady-state offset
• Derivative: Predicts error rate of change, providing damping

Control output selects actuator based on altitude: if h > 30km, command RCS thrusters; if 10km < h < 30km, blend RCS and grid fins; if h < 10km, command grid fins only.

Performance Analysis

The simulation tracks key performance metrics across the 40km descent:

• Lateral Deviation: Maximum off-axis displacement from target trajectory (< 50m achieved)
• Fuel Consumption: RCS propellant mass expended (directly proportional to thruster firing duration)
• Control Saturation: Percentage of timesteps where actuators reach maximum deflection (indicates under-actuated regions)
• Settling Time: Time required to converge within ±1° of target attitude after disturbance injection

Corrected physics implementation achieved 100× performance improvement over initial version—reducing 10-minute simulation runtime to 6 seconds.

Key Insights

Mode Transition Challenges: Abrupt switching at altitude thresholds causes control discontinuities. Smooth blending (weighted average of RCS and aerodynamic control based on dynamic pressure) reduces oscillations by 60%.

Aerodynamic Instability: Grid fins destabilize the vehicle if center of pressure shifts ahead of center of mass—requiring continuous corrective torque. Properly tuned derivative gain (damping) prevents divergent oscillations.

RCS Fuel Budget: Aggressive PID tuning (high gains) minimizes settling time but exhausts propellant reserves. Trade-off: accept 2× longer settling in exchange for 40% fuel savings.

Technical Resources

Full simulation code, Jupyter notebooks, and visualizations available on GitHub:

→ View on GitHub

Repository includes both original (sim.ipynb) and corrected (sim_corrected.ipynb) implementations with detailed physics error documentation.