StarIO: A Lightweight Inertial Odometry for Nonlinear Motion

TL;DR

StarIO tackles drift in inertial odometry under nonlinear motion by projecting IMU signals ${\mathbf{X}} \in \mathbb{R}^{C\times L}$ with $C=6$ into a high-dimensional implicit nonlinear feature space via the Star Operation, enabling richer motion representations processed by a lightweight Transformer-based encoder with Dual-Wing Attention and MSGCU. The approach yields accurate displacement estimates by averaging velocity over the window and integrating, trained with $\text{MSE}$. Empirical results across six pedestrian-inertial datasets show substantial improvements in ATE, RTE, and PDE, with RoNIN achieving reductions up to $65.78\%$ in ATE and StarIO offering a favorable accuracy-parameter trade-off. The work demonstrates practical, edge-friendly IO capable of modeling complex nonlinear motion and points to future extensions to non-pedestrian platforms such as ground vehicles and UAVs.

Abstract

Inertial odometry (IO) directly estimates the position of a carrier from inertial sensor measurements and serves as a core technology for the widespread deployment of consumer grade localization systems. While existing IO methods can accurately reconstruct simple and near linear motion trajectories, they often fail to account for drift errors caused by complex motion patterns such as turning. This limitation significantly degrades localization accuracy and restricts the applicability of IO systems in real world scenarios. To address these challenges, we propose a lightweight IO framework. Specifically, inertial data is projected into a high dimensional implicit nonlinear feature space using the Star Operation method, enabling the extraction of complex motion features that are typically overlooked. We further introduce a collaborative attention mechanism that jointly models global motion dynamics across both channel and temporal dimensions. In addition, we design Multi Scale Gated Convolution Units to capture fine grained dynamic variations throughout the motion process, thereby enhancing the model's ability to learn rich and expressive motion representations. Extensive experiments demonstrate that our proposed method consistently outperforms SOTA baselines across six widely used inertial datasets. Compared to baseline models on the RoNIN dataset, it achieves reductions in ATE ranging from 2.26% to 65.78%, thereby establishing a new benchmark in the field.

Figures (4)

• Figure 1: Comparison of Absolute Trajectory Error (ATE) for different models on the RoNIN dataset. Points nearer the bottom-left indicate lower trajectory error and fewer parameters, reflecting a superior overall performance-efficiency trade-off.
• Figure 2: The overall architecture of the proposed StarIO, which primarily consists of the Lightweight Transform Block (LTB) featuring the Dual-Wing Star Block (DWSB) and Multi-Scale Gated Convolution Unit (MSGCU).
• Figure 3: Performance comparison of StarIO, R-ResNet, and R-LSTM evaluated on the TLIO dataset. (a) and (b) show the CDF of ATE and RTE, respectively, while subfigure (c) presents a box plot of PDE.
• Figure 4: Sample trajectories from six test datasets. The results compare the performance of the proposed StarIO model against the R-ResNet baseline. Values in parentheses indicate the ATE, RTE, and trajectory length, all measured in meters.