Orientation-free neural network-based bias estimation for low-cost stationary accelerometers

Low-cost micro-electromechanical accelerometers are widely used in navigation, robotics, and consumer devices for motion sensing and position estimation. However, their performance is often degraded by bias errors. To eliminate deterministic bias terms, a calibration procedure is applied under stationary conditions. It requires accelerometer leveling or complex orientation-dependent calibration procedures. To overcome those requirements, in this paper, we present a model-free learning-based calibration method that estimates accelerometer bias under stationary conditions, without requiring knowledge of the sensor orientation and without requiring structured multi-position calibration sequences during stationary calibration. The proposed orientation-free neural network-based bias estimation model (OFBENet) provides a fast, practical, and scalable solution suitable for rapid stationary calibration. Across a 13.39-h dataset collected from six low-cost accelerometers, OFBENet consistently outperformed traditional calibration techniques, achieving root mean squared error (RMSE) reductions of more than 78% in gravity-aligned conditions and 77–87% in rotated configurations, with RMSE values as low as 0.04 m/s². On a broader scale, this work contributes to the advancement of accurate calibration methods in orientation-free scenarios. The ability to estimate accelerometer bias without controlled orientations or external equipment reduces both the calibration cost and the operational complexity. As a result, the proposed method is a step toward reliable bias estimation for low-cost accelerometers in applications such as robotics and navigation, particularly in scenarios requiring rapid initialization or in-field calibration. As a consequence, it supports the reliable use of low-cost inertial sensors in various scientific and industrial applications and eliminates the need for level calibration.