Spiking World Model with Multi-Compartment Neurons for Model-based Reinforcement Learning

Yinqian Sun^1,4, Feifei Zhao^1,3,4, Mingyang Lyu^1,2, Yi Zeng^1,2,3,4,

¹Brain-inspired Cognitive AI Lab, Institute of Automation, Chinese Academy of Sciences, ²School of Artificial Intelligence, University of Chinese Academy of Sciences ³State Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology ⁴Long-term AI

💡 Introduction

Spiking-WM is a brain-inspired spiking world model for model-based reinforcement learning that introduces multi-compartment neurons (MCNs) to equip SNNs with long-term temporal memory. Inspired by nonlinear dendritic integration in biological neurons, Spiking-WM integrates a spiking state-space model, a spiking encoder, and a spiking policy network to enable end-to-end planning and decision-making. Experiments on the DeepMind Control Suite demonstrate that Spiking-WM outperforms existing SNN-based approaches and achieves performance comparable to GRU-based ANN world models, while evaluations on long-sequence speech benchmarks (SHD, TIMIT, and LibriSpeech 100h) further confirm its superior capability for modeling long-range temporal dependencies.

📢 News

[2025-12-12] Spiking-WM has been published online in Proceedings of the National Academy of Sciences (PNAS).

Table of Contents

• 💡 Introduction
• 📢 News
• Table of Contents
• 🛠️ Installation
• 🚀 Training
• 📜 Citing
• 🙏 Acknowledgement

🛠️ Installation

Create the required environment through the following steps:

conda env create -f environment.yml
conda activate spiking-wm
pip install -r requirements.txt

🚀 Training

bash scripts/train.sh

📜 Citing

If you find Spiking-WM is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry:

@article{sun2025spiking,
author = {Yinqian Sun  and Feifei Zhao  and Mingyang Lyu  and Yi Zeng },
title = {Spiking world model with multicompartment neurons for model-based reinforcement learning},
journal = {Proceedings of the National Academy of Sciences},
volume = {122},
number = {50},
pages = {e2513319122},
year = {2025},
doi = {10.1073/pnas.2513319122},
URL = {https://www.pnas.org/doi/abs/10.1073/pnas.2513319122},
eprint = {https://www.pnas.org/doi/pdf/10.1073/pnas.2513319122},
}

The model of this research is one of the core and part of BrainCog Embot.BrainCog Embot is an Embodied AI platform under the Brain-inspired Cognitive Intelligence Engine (BrainCog) framework, which is an open-source Brain-inspired AI platform based on Spiking Neural Network.

@article{zeng2023braincog,
title={BrainCog: A spiking neural network based, brain-inspired cognitive intelligence engine for brain-inspired AI and brain simulation},
author={Zeng, Yi and Zhao, Dongcheng and Zhao, Feifei and Shen, Guobin and Dong, Yiting and Lu, Enmeng and Zhang, Qian and Sun, Yinqian and Liang, Qian and Zhao, Yuxuan and others},
journal={Patterns},
volume={4},
number={8},
year={2023},
publisher={Elsevier}
}

🙏 Acknowledgement

Our work is primarily based on the following codebases:dreamerv3-torch. We are sincerely grateful for their work.