The field of inertial localization and odometry is rapidly advancing with a focus on improving accuracy and efficiency. Recent developments have seen the proposal of novel network architectures that integrate both frequency-domain and time-domain information to effectively model long-term dependency in inertial measurement unit data. Other approaches have introduced lightweight frameworks that project inertial data into high-dimensional implicit nonlinear feature spaces, enabling the extraction of complex motion features. Additionally, there has been a growing interest in incorporating physics-informed neural networks and adaptive filtering techniques to enhance robustness and accuracy. These innovative methods have demonstrated significant improvements in localization accuracy and have the potential to enable widespread deployment of consumer-grade localization systems. Noteworthy papers include: RepILN, which proposes a reparameterized inertial localization network that achieves a favorable trade-off between accuracy and model compactness. FTIN, which introduces a frequency-time integration network that captures long-term dependency in inertial measurement unit data and reduces absolute trajectory error by 43.0%. DWSFormer, which presents a lightweight inertial odometry network that consistently outperforms state-of-the-art baselines across six widely used inertial datasets.