Fei Song

Bio

I received my Ph.D. from the Shenyang Institute of Automation, Chinese Academy of Sciences. During my doctoral studies, I was advised by Prof. Bailu Si, who moved to Beijing Normal University during my Ph.D. period.

My research broadly focuses on spatial cognition, planning, and computational modeling, exploring how insights from biological intelligence can inspire principles for artificial intelligence. I am particularly interested in the mechanisms underlying spatial navigation, goal-directed behavior, and planning, and have previously developed computational models and algorithmic frameworks for physical-space navigation and path planning.

I am currently a Postdoctoral Researcher in the School of Psychology at Shanghai Jiao Tong University, where I work with Prof. Ruyuan Zhang. My current research focuses on human cognition and planning, aiming to bridge computational algorithms with cognitive theories to better understand the representational and inferential processes that underlie intelligent behavior.

News

• [10/2025]  Our collaborative work 📐 The Principle of Isomorphism: A Theory of Population Activity in Grid Cells and Beyond 🧩 is now available on arXiv.
• [08/2025]  I officially started my postdoctoral research position at the School of Psychology, Shanghai Jiao Tong University, joining Prof. Ruyuan Zhang’s research group.
• [06/2025]  I successfully obtained my Ph.D. degree! 🎓 I am deeply grateful for the rigorous academic training at the Shenyang Institute of Automation, and for the inspiring research atmosphere at the School of Systems Science, Beijing Normal University, both of which shaped my doctoral journey. A heartfelt thanks to my mentor, Prof. Bailu Si 🙏, for his guidance, encouragement, and long-term support. I also cherish all the wonderful friends, collaborators, and labmates I met at BNU 🤝🌟 — our discussions, shared ideas, and the time we spent together made these years truly memorable.
• [05/2025]  Our work 🤖 An Improved Artificial Potential Field Method with Distributed Representation and Scale-Invariant Path Planning 🔍 has been accepted by IEEE Transactions on Cognitive and Developmental Systems.
• [04/2025]  Our paper 🧭 A hippocampal navigation model through hierarchical memory organization 🧠 has been accepted in Cognitive Neurodynamics.

Publications

An Improved Artificial Potential Field Method with Distributed Representation and Scale-Invariant Path Planning
Fei Song, Yuxiu Shao, Dengyao Jiang, Ziyu Ren, Fengzhen Tang, Yandong Tang, Bailu Si

IEEE Transactions on Cognitive and Developmental Systems, 2025

Inspired by distributed encoding mechanisms in neural receptive fields, this work proposes the Neuro-Receptive Field Planner (NRF) — a novel reformulation of the Artificial Potential Field (APF) method that replaces traditional scalar-force definitions with distributed attraction and repulsion representations, combined with an automatic adaptive adjustment mechanism. NRF effectively decouples scale parameters, improves interpretability, and demonstrates robust, scale-invariant path planning performance across static and dynamic environments, achieving the lowest average CV score among all baselines.

Paper Cite

@ARTICLE{11091598, 
  author={Song, Fei and Shao, Yuxiu and Jiang, Dengyao and Ren, Ziyu and Tang, Fengzhen and Tang, Yandong and Si, Bailu},
  journal={IEEE Transactions on Cognitive and Developmental Systems}, 
  title={An Improved Artificial Potential Field Method with Distributed Representation and Scale-Invariant Path Planning}, 
  year={2025},
  volume={},
  number={},
  pages={1-15},
  keywords={Force;Vectors;Encoding;Path planning;Neurons;Heuristic algorithms;Artificial intelligence;Silicon;Planning;Tuning;Path planning;Navigation;Artificial Potential Field Method;Distributed Representation;Brain-inspired Algorithm},
  doi={10.1109/TCDS.2025.3592082}
}

A Hippocampal Navigation Model Through Hierarchical Memory Organization
Fei Song, Jinyu Li, Fenzhen Tang, Yandong Tang, Bailu Si

Cognitive Neurodynamics, 2025

Inspired by the hippocampal microcircuit and the ventral–dorsal architecture, this work develops a unified hierarchical memory-based navigation model that dynamically transitions between vector navigation and episodic memory-guided path planning. Leveraging grid-cell decoding, multi-scale episodic memory storage, and border-cell–based movement constraints, the model reliably replicates complex navigation behaviors including goal-directed planning, efficient shortcut discovery, and robust direction selection—even under simulated hippocampal lesions.

Paper Cite

        @article{song2025hippocampal,
        title={A hippocampal navigation model through hierarchical memory organization},
        author={Song, Fei and Li, Jinyu and Tang, Fenzhen and Tang, Yandong and Si, Bailu},
        journal={Cognitive Neurodynamics},
        volume={19},
        number={1},
        pages={103},
        year={2025},
        publisher={Springer},
        doi={10.1007/s11571-025-10254-w}
        }
        

Spatiotemporal Dual-Stream Network for Visual Odometry
Chang Xu, Taiping Zeng, Yifan Luo, Fei Song, Bailu Si

IEEE Robotics and Automation Letters (RA-L), 2025

We propose a novel monocular visual odometry framework, the Spatiotemporal Dual-Stream Network (STDN-VO), consisting of two parallel branches designed to extract spatial global context and sequential temporal dependencies from image sequences. Experiments on the KITTI dataset demonstrate superior performance over existing deep learning-based VO methods, highlighting robustness and improved pose estimation accuracy.

Paper Cite

@ARTICLE{10897744,
  author={Xu, Chang and Zeng, Taiping and Luo, Yifan and Song, Fei and Si, Bailu},
  journal={IEEE Robotics and Automation Letters},
  title={Spatiotemporal Dual-Stream Network for Visual Odometry},
  year={2025},
  volume={10},
  number={4},
  pages={3867-3874},
  keywords={Feature extraction;Visualization;Correlation;Transformers;Spatiotemporal phenomena;Pose estimation;Data mining;Visual odometry;Deep learning;Odometry;Monocular visual odometry;dual-stream network;deep learning},
  doi={10.1109/LRA.2025.3544521}}