Learning Attractor Dynamics via Reservoir Computing for Efficient and Robust Path Integration

The code for "Learning Attractor Dynamics via Reservoir Computing for Efficient and Robust Path Integration".

BibTeX Citation

@inproceedings{zhang2026learning,
  title = {Learning Attractor Dynamics via Reservoir Computing for Efficient and Robust Path Integration},
  booktitle = {2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  author = {Zhang, Youdong and He, Xu and Meng, Xiaolin and Mo, Lingfei and An, Xiangdong},
  year = 2026
}

Introduction

This repository contains the code for the paper "Learning Attractor Dynamics via Reservoir Computing for Efficient and Robust Path Integration". The code is organized into several modules, including data, model training, and inference. The main components of the code are as follows:

• data/: This directory contains the data for training and testing the model.
• model/: This directory contains the implementation of the reservoir computing model, including the training and inference algorithms.
• utils/: This directory contains utility functions for data processing, model evaluation, and visualization.
• main.py: This is the main script for path integration, which includes the time cost evaluation and visualization of the results.
• requirements.txt: This file lists the required Python packages for running the code. It can be used to set up a virtual environment with the necessary dependencies.

Usage

To run the code, follow these steps:

1. Clone the repository and navigate to the project directory.
2. Set up a virtual environment and install the required packages using the command:

   pip install -r requirements.txt
   

3. Run the main script to perform path integration and evaluate the time cost:

    python main.py
   

Results

Comparison of trajectories between NeuroSLAM and the proposed method on (a) KITTI-00, (b) KITTI-07, and (c) the self-collected dataset. The trajectories are aligned with the ground truth using rigid transformations and uniform scaling to preserve topological consistency.

Comparison of the ATE

Comparison of Average Processing Time per Frame on PC and Edge Platforms

License

This project is licensed under the Apache License. See the LICENSE file for details.