No Need for Landmarks: An Embodied Neural Controller for Robust Insect-Like Navigation Behaviors

Area of Science:

• Robotics
• Artificial Intelligence
• Neuroscience

Background:

• Bayesian filters are explored for spatial cell functions and self-localization.
• Reproducing insect-like navigation behaviors (e.g., home searching) with Bayesian filters is challenging.
• Existing methods often rely on global positioning or landmark guidance.

Purpose of the Study:

• To propose an embodied neural controller for an insect-like robot (AMOS) capable of self-localization, foraging, backward homing, and home searching.
• To achieve robust insect-like navigation behaviors using only self-localization and heading perception.
• To compare the controller's performance against conventional methods under sensory noise.

Main Methods:

• Developed an embodied neural controller with navigation and locomotion modules for an advanced mobility sensor (AMOS)-driven robot.
• The navigation module integrates Bayesian self-localization and goal-directed control.
• The locomotion module coordinates AMOS's 18 joints for robust movement.

Main Results:

• The proposed controller enabled AMOS to perform robust foraging, backward homing, and home searching.
• AMOS demonstrated resilience to various levels of sensory noise, outperforming conventional controllers.
• The robot exhibited spiral searching patterns similar to real insects, without global positioning or landmarks.
• Successful real-time indoor and outdoor navigation was achieved.

Conclusions:

• The embodied neural controller facilitates robust insect-like navigation in robots.
• The system relies solely on self-localization and heading perception, eliminating the need for landmarks or cognitive maps.
• The controller's ability to replicate insect searching patterns highlights its potential for bio-inspired robotics.