First-Class BEng (Hons) Aerospace Engineering · 90%

Autonomous Flight Control
using Reinforcement Learning

Reinforcement Learning-Based Control for High-Performance
Aircraft Manoeuvring in Nonlinear Flight Regimes

PID — Classical
SAC — Reinforcement Learning

Fully autonomous 360° roll. Landing gear visible in the renders is for visual reference only — not aerodynamically modelled in the simulation.

90% Dissertation Grade
PID · LQR
PPO · SAC Controllers
96% Altitude Reduction
R²0.99 Predictive Model
150 Training Episodes

Built a Python-based 6-DoF F-16 simulation and evaluated PID, LQR, PPO and SAC controllers under randomised disturbances. SAC reduced altitude loss during an autonomous 360° roll by 96% relative to PID across 50 independently seeded evaluations. A predictive model also achieved R²=0.99, using the first two seconds of flight data to estimate final controller performance.

Figure 19 — Altitude loss across manoeuvres

Altitude loss across manoeuvres — PID, PPO, SAC
Controller performance across the evaluated manoeuvres. SAC consistently achieved the lowest altitude loss and the greatest robustness, with a 96% reduction during the autonomous 360° roll compared with the PID baseline.

System Architecture

Python
Simulation & training environment
JSBSim
Six-degree-of-freedom physics engine
Gymnasium Environment
State · Action · Reward
Controllers
PID
LQR
PPO
SAC
Stable-Baselines3
RL training & evaluation
Performance Metrics
Altitude · Tracking error · Robustness

Flight Dynamics

Control Systems

Software

Dissertation

Autonomous Flight Control using Reinforcement Learning

BEng Aerospace Engineering · First Class (90%) · 44 pages · PDF

View PDF ↗︎

Opportunities

BEng Aerospace Engineering graduate open to roles in engineering, research, and adjacent technical fields. Feel free to get in touch regarding CV or opportunities.

Get in touch ↗︎