🚀 Real-EqNIO: Real-Time Equivariant Neural Inertial Odometry

Real-EqNIO is a real-time Neural Inertial Odometry (NIO) system designed specifically for physical robot deployments. By leveraging the power of equivariant neural networks, it ensures consistent and accurate 6-DoF state estimation across various transformations using purely inertial (IMU) data, entirely independent of visual inputs.

📢 Acknowledgement & Foundation

This repository focuses on the real-world deployment and real-time inference of our foundational research. The core IMU pre-integration and equivariant network formulations are heavily built upon our previous work. For the standalone offline network architecture, training pipelines, and datasets, please refer to:

• 🔗 IMU Equivariant Learning: A yaw-Equivariant Network-Coupled Inertial Odometry.

⚙️ System Requirements

To run Real-EqNIO on a physical robot, the following hardware and software setups are recommended:

• Compute Environment: Edge computing devices suitable for real-time deep learning inference (e.g., NVIDIA Jetson series, Intel NUC).
• Sensor: A 6-DoF IMU providing high-frequency data streams (100Hz - 400Hz+).
• Middleware: ROS / ROS 2 for asynchronous sensor data handling and odometry publishing.