In this work, a novel neural-adaptive nonlinear delay stochastic filter is designed to address the attitude estimation problem. This filter is represented using the special orthogonal group SO ( 3 ) , and employs low-cost sensors’ units. Specifically, the Brownian motion is introduced to characterize the noise that maps the system dynamics to stochastic differential equations (SDEs), ensuring that the attitude estimation problem can be analyzed in a stochastic sense. Neural networks (NNs) are employed to design an attitude filter that accounts for sensor measurement delay and noise by incorporating a Lyapunov function. This stochastic filter design ensures that the closed-loop system is almost semi-globally uniformly ultimately bounded (SGUUB) in the mean square sense. Finally, simulations to verify the efficiency of the proposed attitude filter.