CKANIO: Learnable Chebyshev Polynomials for Inertial Odometry

TL;DR

CKANIO addresses drift in inertial odometry by modeling nonlinear IMU motion with a learnable Chebyshev Kolmogorov–Arnold Network (KAN) and by augmenting contextual modeling with an efficient kernel-based self-attention module. The approach pairs a 1D Chebyshev KAN-based residual block (ResCKAN) with EKSA, enabling linear-time attention and robust feature extraction from IMU signals, followed by velocity prediction via an MLP. It is claimed to be the first interpretable KAN application in IO and demonstrates consistent gains across five public datasets, including gravity-aware analyses and ablations showing contributions from both ResCKAN and EKSA. The work offers a new architectural direction for IO that better captures nonlinear dynamics and may extend to broader IMU signal regimes beyond pedestrian scenarios.

Abstract

Inertial odometry (IO) relies exclusively on signals from an inertial measurement unit (IMU) for localization and offers a promising avenue for consumer grade positioning. However, accurate modeling of the nonlinear motion patterns present in IMU signals remains the principal limitation on IO accuracy. To address this challenge, we propose CKANIO, an IO framework that integrates Chebyshev based Kolmogorov-Arnold Networks (Chebyshev KAN). Specifically, we design a novel residual architecture that leverages the nonlinear approximation capabilities of Chebyshev polynomials within the KAN framework to more effectively model the complex motion characteristics inherent in IMU signals. To the best of our knowledge, this work represents the first application of an interpretable KAN model to IO. Experimental results on five publicly available datasets demonstrate the effectiveness of CKANIO.

Figures (4)

• Figure 1: Illustration of the processing pipeline for data-driven IO.
• Figure 2: Network Architecture Diagram of CKANIO.
• Figure 3: PDE comparison between CKANIO and three baseline models from RoNIN on the RoNIN dataset.
• Figure 4: Visual comparison of CKANIO and RoNIN ResNet on representative test trajectories from multiple datasets.