Systems Theoretic Process Analysis of a Run Time Assured Neural Network Control System

TL;DR

This paper applies Systems Theoretic Accident Models and Processes (STAMP) and System Theoretic Process Analysis (STPA) to identify safety constraints and develop Run Time Assurance for neural network control systems in autonomous flight, ensuring collision avoidance.

Contribution

It is the first to apply STAMP and STPA to a neural network control system bounded by Run Time Assurance in autonomous flight scenarios.

Findings

Identified safety constraints and hazards for NNCS in autonomous formation flight.

Developed safety requirements to mitigate risks using STPA.

Demonstrated the application of systems theoretic methods to neural network control systems.

Abstract

This research considers the problem of identifying safety constraints and developing Run Time Assurance (RTA) for Deep Reinforcement Learning (RL) Tactical Autopilots that use neural network control systems (NNCS). This research studies a specific use case of an NNCS performing autonomous formation flight while an RTA system provides collision avoidance and geofence assurances. First, Systems Theoretic Accident Models and Processes (STAMP) is applied to identify accidents, hazards, and safety constraints as well as define a functional control system block diagram of the ground station, manned flight lead, and surrogate unmanned wingman. Then, Systems Theoretic Process Analysis (STPA) is applied to the interactions of the the ground station, manned flight lead, surrogate unmanned wingman, and internal elements of the wingman aircraft to identify unsafe control actions, scenarios leading to each, and safety requirements to mitigate risks. This research is the first application of STAMP and STPA to an NNCS bounded by RTA.