Automated Aircraft Stall Recovery using Reinforcement Learning and Supervised Learning Techniques

Despite the on-board automation and protection systems of modern commercial aircraft, aerodynamic stall events are still a possible occurrence. This paper proposes Machine Learning algorithms – based on Reinforcement Learning and Supervised Learning – to automatically recover an aircraft from two types of aerodynamic stall: unaccelerated wings level (1G) stall and a stall during a turn. The algorithms were tested by exposing them to 105 simulated stall scenarios with different initial conditions (including altitude, bank angle and wind speed) and an acceptable stall recovery was achieved in 85.7% of the test cases. The overall recovery time increased with an increase in altitude, with the best and worst recovery times obtained at 10,000ft and 30,000ft respectively. Further work will focus on improving the performance of the algorithms such as by reducing the time to recover from a stall, decreasing the altitude loss and training the algorithms over a larger range of altitudes, up to cruise level.