A biologically inspired neural network for optic flow-based reactive navigation with dual depth encoding.

Area of Science:

• Neuroscience
• Robotics
• Computer Vision

Background:

• Optic flow is crucial for navigation in animals and humans.
• Biological neural networks for optic flow-based navigation are underexplored.
• Existing methods often lack biological plausibility or robustness in complex environments.

Purpose of the Study:

• To propose a biologically plausible neural network model for optic flow-based reactive navigation.
• To enhance heading sensitivity and obstacle localization for improved navigation.
• To validate the model's performance in simulations and real-world experiments.

Main Methods:

• Developed a neural network model with primary visual cortex, higher-order cortex, and cerebellum.
• Incorporated a feedback inhibitory pathway (V1 layer VI to layer IV) for enhanced heading sensitivity.
• Utilized a dual encoding strategy combining optic flow with depth maps for precise obstacle localization.

Main Results:

• The model successfully enabled collision-free navigation in diverse scenarios.
• Demonstrated superior performance compared to classical optic flow balance strategies in complex environments.
• Validated the model's effectiveness through simulations and real-world experiments with an intelligent vehicle.

Conclusions:

• Biologically inspired neural networks offer a feasible solution for visual reactive navigation in autonomous agents.
• The proposed model enhances navigation capabilities by integrating optic flow, depth perception, and biologically plausible neural mechanisms.
• This work opens new avenues for developing more sophisticated and adaptable autonomous navigation systems.