A fast, three-layer neural network for path finding

Area of Science:

• Artificial Intelligence
• Robotics
• Computational Neuroscience

Background:

• Path-planning is crucial for autonomous systems.
• Existing algorithms like diffusion and potential fields have limitations.
• Artificial neural networks offer potential for complex navigation tasks.

Purpose of the Study:

• To develop an improved path-planning algorithm using a three-layer artificial neural network.
• To overcome drawbacks of classical diffusion and potential field methods.
• To achieve robust navigation with obstacle avoidance and smooth path generation.

Main Methods:

• A three-layer artificial neural network with local rules and recurrent connections.
• Modification of a diffusion process with nonlinear transformation for wave-like propagation.
• Integration of 'obstacle potentials' to manage proximity to obstacles.
• Utilizing coarse coding for spatial interpolation and smooth path formation.

Main Results:

• The algorithm successfully plans paths by adapting diffusion processes.
• It exhibits wave propagation properties, overcoming resolution limitations.
• Obstacle potentials allow for adjustable safety margins, enhancing safety.
• The approach combines benefits of diffusion, wave propagation, and potential fields, avoiding local minima.

Conclusions:

• The proposed artificial neural network-based path-planning algorithm offers a robust solution.
• It effectively integrates multiple navigation strategies for improved performance.
• The method demonstrates potential for autonomous systems requiring safe and efficient path planning.