Technology Category
Intelligent Robots: State EstimationComputer Vision: Motion & TrackingMachine Learning: Imitation Learning & Inverse Reinforcement Learning
ResKACNNet: A Residual ChebyKAN Network for Inertial Odometry
🤖 AI Summary
To address the limitations of CNNs in modeling nonlinear motion characteristics and long-range temporal dependencies in IMU data, this paper proposes Residual ChebyKAN Inertial Odometry, a novel neural architecture. Methodologically: (1) Chebyshev polynomials are integrated into Kolmogorov–Arnold Networks (KANs) to enhance their nonlinear approximation capability; (2) an Efficient Kernel-based Self-Attention (EKSA) module is designed to improve long-horizon sequential modeling; (3) gravitational acceleration components are explicitly removed from raw accelerometer measurements to boost pose estimation accuracy. The model is trained and evaluated end-to-end on standard benchmarks—including RIDI and RoNIN—achieving significant reductions in absolute trajectory error (ATE) ranging from 3.79% to 42.32%. To foster reproducibility and community advancement, the authors publicly release both preprocessed datasets and source code, establishing a new baseline for learning-based inertial navigation research.
Technology Category
Application Category
📝 Abstract
Inertial Measurement Unit (IMU) has become a key technology for achieving low-cost and precise positioning. However, traditional CNN-based inertial positioning methods struggle to capture the nonlinear motion characteristics and long-term dependencies in IMU data. To address this limitation, we propose a novel inertial positioning network with a generic backbone called ResChebyKAN, which leverages the nonlinear approximation capabilities of Chebyshev polynomials to model complex motion patterns. Additionally, we introduce an Efficient Kernel-based Self-Attention (EKSA) module to effectively capture contextual information and enhance long-term dependency modeling. Experimental results on public datasets (e.g., RIDI, RoNIN, RNIN-VIO, OxIOD, IMUNet, and TLIO) demonstrate that our method reduces the absolute trajectory error by 3.79% to 42.32% compared to existing benchmark methods. Furthermore, we release a preprocessed dataset and empirically show that removing the gravity component from acceleration data significantly improves inertial positioning performance.
Problem
Research questions and friction points this paper is trying to address.
Improves IMU-based positioning accuracy using Chebyshev polynomials
Captures long-term dependencies in IMU data via EKSA module
Enhances performance by removing gravity from acceleration data
Innovation
Methods, ideas, or system contributions that make the work stand out.
ResChebyKAN leverages Chebyshev polynomials for motion modeling
EKSA module enhances long-term dependency modeling
Gravity component removal improves positioning performance
🔎 Similar Papers
No similar papers found.
