Neuromorphic control of stepping pattern generation: a dynamic model with analog circuit implementation.

Area of Science:

• Robotics and Neuromorphic Engineering
• Biomimetic Locomotion Systems

Background:

• Animals like stick insects exhibit adaptive locomotion on complex terrains by modulating stepping patterns.
• Existing models for rhythmic pattern generation include coupled Matsuoka and resonate-and-fire neuron models.

Purpose of the Study:

• To present a nonlinear oscillation model as a neuromorphic central pattern generator (CPG) for rhythmic stepping pattern generation.
• To implement a mixed-signal integrated circuit of the dynamic CPG model for actuating motoneurons with adjustable parameters.
• To demonstrate the potential for adaptive walking on complex terrains using a three-jointed arthropod leg driven by CPG models.

Main Methods:

• Developed a nonlinear oscillation model inspired by Matsuoka and resonate-and-fire neuron models for CPG function.
• Designed and implemented a novel mixed-signal integrated circuit of the dynamic CPG model.
• Utilized three identical CPG models to drive a three-degrees-of-freedom arthropod leg, focusing on adjustable frequencies, duty cycles, and phase lags.

Main Results:

• The implemented CPG circuit demonstrated equivalent output performance to the dynamic model, with adjustable frequency and duty cycle.
• The three-jointed leg, driven by CPGs with appropriate parameter settings, showed potential for adaptive stepping on complex terrain.
• Adaptation mechanisms were linked to circuit parameters influenced by higher-level neural control and lower-level sensory feedback.

Conclusions:

• A neuromorphic CPG model based on nonlinear oscillations can effectively generate adjustable rhythmic stepping patterns.
• The mixed-signal integrated circuit implementation provides a viable hardware platform for biomimetic locomotion.
• The proposed system holds promise for developing robots capable of adaptive locomotion in unstructured environments.