Deriving neural network controllers from neuro-biological data: implementation of a single-leg stick insect...

Area of Science:

• Robotics
• Computational Neuroscience
• Biomimetics

Background:

• Biomimetic robots often draw inspiration from insect locomotion for enhanced agility.
• Decentralized control systems are crucial for replicating complex biological movement patterns.

Purpose of the Study:

• To develop and evaluate modular recurrent neural network controllers for individual legs of a hexapod robot.
• To investigate the feasibility of using neuro-biological data from stick insects to design robot controllers.
• To assess the robustness and adaptability of these controllers under various environmental conditions.

Main Methods:

• Derivation of decentralized, sensori-driven neuro-controllers from stick insect neuro-biological data.
• Implementation of modular recurrent neural network architectures for single robot legs.
• Parameter tuning via hand-tuning and evolutionary algorithms.
• Simulation of robot locomotion and comparison with biological data.

Main Results:

• Achieved qualitatively similar behaviors in robot and insect simulations using identical controller structures.
• Demonstrated robust swing and stance control under perturbing conditions like varying ground heights and slopes.
• Observed behavioral adaptations, such as speed variation, through neural parameter changes and environmental coupling.
• Simulated walking behavior closely matched biological data, including force profiles and trajectory patterns.

Conclusions:

• The developed single-leg controllers are suitable modules for hexapod locomotion control.
• These modular controllers can bridge morphological and behavioral approaches in robotics.
• The findings support the potential of neuro-biologically inspired control for advanced robotic systems.