Biomimetic FPGA-based spatial navigation model with grid cells and place cells.

Area of Science:

• Neuroscience
• Robotics
• Computer Engineering

Background:

• Mammalian spatial navigation relies on specialized neurons for location, speed, and direction, with subsequent convergence for path integration.
• Existing navigation systems often require multiple sensors and significant computational resources.

Purpose of the Study:

• To implement a biomimetic, multi-modular neural architecture for spatial navigation.
• To create a hardware-based system for precise and efficient trajectory estimation.

Main Methods:

• Designed a feed-forward, trimodular architecture with grid-cell, place-cell, and decoding modules.
• Implemented the biomimetic structure on a Zynq Ultrascale+ field-programmable gate array (FPGA).
• Optimized modules for performance and resource utilization.

Main Results:

• The system accurately estimated navigational trajectories with low FPGA resource utilization (2.92% LUT).
• Achieved real-time processing (32s for 100k samples) significantly faster than CPU-based methods.
• Demonstrated stability, reliability, and reconfigurability.

Conclusions:

• The biomimetic digital spatial navigation system offers a stable, efficient, and real-time solution.
• This FPGA-based implementation is suitable for autonomous-robotic navigation without additional sensors.
• The study validates the potential of neuromorphic engineering for advanced navigation tasks.