Collective motion model inspired by fish school based on deep attention mechanism

Area of Science:

• Computational neuroscience
• Robotics
• Collective behavior

Background:

• Modeling social interactions in collective systems is challenging.
• Biological collective intelligence offers inspiration for artificial systems like swarm robotics.
• Understanding information processing in schooling fish can inform agent-based modeling.

Purpose of the Study:

• To propose a deep attention network model inspired by biological Hard Attention mechanisms for collective motion.
• To train and validate the model using collective movement data from rummy-nose tetra fish.
• To analyze the model's ability to handle sparse information and its scalability.

Main Methods:

• Developed a deep attention network incorporating Hard Attention principles, limiting agent focus to 1-2 neighbors.
• Trained the model on collective movement data from five rummy-nose tetra fish (Hemigrammus rhodostomus).
• Evaluated model performance on information decoupling, scalability across group sizes, and real-world robotic experiments.

Main Results:

• The Hard Attention Model successfully mimics information processing observed in fish schooling.
• The model demonstrates robust metrics for decoupling sparse information crucial for collective movement.
• Excellent scalability was observed across different group sizes, with successful validation in simulations and on physical robots.

Conclusions:

• The proposed Hard Attention Model provides a powerful tool for analyzing multi-level behaviors in complex systems.
• Findings offer significant insights for the distributed control of swarm robotics by leveraging biological principles.
• The model effectively processes information from limited neighbors, aligning with biological observations.