CASA-KalmanNet self-adapts to system changes with neural change detection

A team from academia and industry has unveiled CASA-KalmanNet, a change-aware self-adaptive AI-aided Kalman filter designed to maintain reliable state estimation in dynamical systems plagued by model mismatches, unknown noise statistics, and temporal variations. Traditional AI-aided filters like KalmanNet leverage deep learning but remain vulnerable to distribution shifts and lack autonomous adaptation. CASA-KalmanNet introduces CPDNet, a dedicated neural module that continuously monitors interpretable internal features of KalmanNet, providing soft indicators of reliability degradation. These indicators dynamically regulate an online learning process, enabling data-efficient and timely adaptation to both abrupt and gradual changes in the system—without requiring additional state labels from the changed regime.

Numerical experiments on linear and nonlinear state-space models demonstrate that CASA-KalmanNet consistently outperforms existing learning-based filters under model mismatch while approaching the accuracy of optimal classical methods that assume full domain knowledge. The framework's ability to autonomously detect and adapt to changes in real-time positions it as a significant advancement for applications in navigation, robotics, and autonomous systems where sensor conditions or dynamics shift unpredictably. The paper, published on arXiv, includes 15 pages and 11 figures, with code expected to follow.