A Brain-Inspired Approach for Collision-Free Movement Planning in the Small Operational Space.

Area of Science:

• Robotics and Neuroscience
• Computational Intelligence
• Biologically Inspired Systems

Background:

• Precise control of robotic systems in confined spaces is challenging due to the need for careful manipulation of fragile objects.
• Existing visual guidance systems may lack the sophisticated processing required for complex, real-time adjustments in microscale environments.

Purpose of the Study:

• To propose a novel spiking neural network (SNN) architecture for visual guidance in small operational spaces.
• To develop a dual-channel system for achieving collision-free movements by imitating multiple brain regions.
• To enhance robotic manipulation capabilities for fragile objects.

Main Methods:

• Simulated primary and high-level visual cortices for feature extraction and distance measurement.
• Emulated prefrontal cortex using liquid state machines for trajectory prediction and a feedforward network for movement planning.
• Incorporated a cerebellar network for subconscious reactive control.
• Implemented two distinct pathways: visual-servo control and collision-avoidance.
• Utilized supervised learning for initial training and reinforcement learning for synaptic refinement.

Main Results:

• The proposed SNN structure effectively imitates brain region functions for visual guiding.
• The dual-channel approach successfully achieves collision-free movements in small operational spaces.
• Both simulation and experimental results validate the efficacy of the proposed method for precise robotic control.

Conclusions:

• The developed SNN-based system offers a biologically inspired solution for complex visual guidance tasks.
• The dual-channel architecture provides robust control for handling fragile objects in confined environments.
• This approach advances the integration of neuroscience principles into robotic control systems.