Markov Decision Process Framework for Flight Safety Assessment and Management

Abstract

Loss of control is the most common precursor to aircraft accidents. This paper presents a flight safety assessment and management system aimed at mitigating loss-of-control risks. Nominally, flight safety assessment and management serves as a passive watchdog system. When loss-of-control scenarios are encountered, flight safety assessment and management issues resilient control overrides to restore a safe operational state. This paper formulates flight safety assessment and management as a Markov decision process to account for uncertainties in state evolution and tradeoffs between passive monitoring and safety-based override. To ensure unsafe states are unreachable, probabilistic constraints are incorporated into the Markov decision process formulation. The Markov decision process framework is applied to prevent loss-of-control events during takeoff. An abstract representation of the underlying state space is specified to minimize Markov decision process computational overhead and to facilitate understanding of the resulting policy. Flight safety assessment and management is evaluated in a runway overrun case study motivated by a real-world incident.