Retina-Based Pipe-Like Object Tracking Implemented Through Spiking Neural Network on a Snake Robot

Area of Science:

• Robotics
• Computer Vision
• Neuroscience

Background:

• Bio-inspired snake robots require robust vision systems for environmental exploration.
• Traditional vision modules struggle with image blur from robot motion.
• Neuromorphic vision sensors (NVS) offer high temporal frequency and dynamic range for improved target detection.

Purpose of the Study:

• To develop and evaluate a novel vision-based target tracking system for snake robots.
• To integrate a neuromorphic vision sensor (NVS) with a spiking neural network (SNN) for enhanced tracking capabilities.
• To enable autonomous locomotion and pipe-like object tracking in snake robots.

Main Methods:

• An NVS was employed to capture visual data as an asynchronous event stream.
• A spiking neural network (SNN) incorporating Hough Transform was designed for target detection.
• A tracking framework combined NVS data with joint position sensor information for motion analysis.
• Simulations were conducted to validate the proposed framework and tracking system.

Main Results:

• The integrated NVS and SNN system successfully achieved pipe-like object tracking on a simulated snake robot.
• The proposed tracking framework demonstrated the autonomous locomotion capabilities of the snake robot.
• Performance comparison showed the SNN model achieved optimal speed on a GPU with synchronous updates.
• The SNN model exhibited higher precision on a CPU using asynchronous updates.

Conclusions:

• The developed NVS- and SNN-based system provides a viable solution for vision-based target tracking in snake robots.
• The framework enhances autonomous exploration capabilities by overcoming limitations of traditional vision modules.
• Computational hardware (CPU vs. GPU) influences the trade-off between speed and precision in SNN-based tracking.