Contraction-Guided Learning-Based Observer for UAV State Estimation Subject to Disturbances

Abstract

This paper proposes a state estimation scheme for nonlinear UAV systems subject to bounded disturbances and unknown dynamics. We introduce a Contracting Learning-Based Observer (CLBO) that integrates a contraction-guided observer with a long short-term memory (LSTM) module. Two learning strategies are developed: adaptive gain learning and additive residual learning, enabling a structured separation between stability-oriented observer design and disturbance compensation. The observer design is guided by contraction-theoretic principles to encourage stable estimation-error dynamics, while the learning module improves robustness against model mismatch and bounded disturbances. Numerical simulations on a quadrotor UAV model demonstrate improved transient and steady-state estimation performance compared with an extended Kalman filter (EKF) under both continuous and impulsive disturbance scenarios.